A111D3 D7DbDb
NAFL INST OF STANDARDS & TECH R.I.C.
A1 11 03070606
Sympos/Ultrasonic tissue c
QC100 .U57 N0.525, 1979 C.2 NBS-PUB-C 19
CO
Q
NBS SPECIAL PUBLICATION
525
U.S. DEPARTMENT OF COMMERCE/National Bureau of Standards
National Bureau of Standards
Library, E-01 Admin. B!dg.
OCT 1 1981
ISICGS
Ultrasonic
Tissue Characterization II
NATIONAL BUREAU OF STANDARDS
The National Bureau of Standards' was established by an act of Congress March 3, 1901. The
Bureau's overall goal is to strengthen and advance the Nation's science and technology and
facilitate their effective application for public benefit. To this end, the Bureau conducts
research and provides: (1) a basis for the Nation's physical measurement system, (2) scientific
and technological services for industry and government, (3) a technical basis for equity in
trade, and (4) technical services to promote public safety. The Bureau's technical work is
performed by the National Measurement Laboratory, the National Engineering Laboratory,
and the Institute for Computer Sciences and Technology.
THE NATIONAL MEASURE\iENT LABORATORY provides the national system of
physical and chemical and materials measurement; cooidinates the system with measurement
systems of other nations and furnishes essential services leading to accurate and uniform
physical and chemical measurement throughout the Nation's scientific community, industry,
and commerce; conducts materials research leading to improved methods of measurement,
standards, and data on the properties of materials needed by industry, commerce, educational
institutions, and Government; provides advisory and research services to other Government
Agencies; develops, produces, and distributes Standard Reference Materials; and provides
calibration services. The Laboratory consists of the following centers:
Absolute Physical Quantities^ — Radiation Research — Thermodynamics and
Molecular Science — Analytical Chemistry — Materials Science.
THE NATIONAL ENGINEERING LABORATORY provides technology and technical
services to users in the public and private sectors to address national needs and to solve
national problems in the public interest; conducts research in engineering and applied science
in support of objectives in these efforts; builds and maintains competence in the necessary
disciplines required to carry out this research and technical service; develops engineering data
and measurement capabilities; provides engineering measurement traceability services;
develops test methods and proposes engineering standards and code changes; develops and
proposes new engineering practices; and develops and improves mechanisms to transfer
results of its research to the utlimate user. The Laboratory consists of the following centers:
Applied Mathematics — Electronics and Electrical Engineering^ — Mechanical
Engineering and Process Technology' — Building Technology — Fire Research —
Consumer Product Technology — Field Methods.
THE INSTITUTE FOR COMPUTER SCIENCES AND TECHNOLOGY conducts
research and provides scientific and technical services to aid Federal Agencies in the selection,
acquisition, application, and use of computer technology to improve effectiveness and
economy in Government operations in accordance with Public Law 89-306 (40 U.S.C. 759),
relevant Executive Orders, and other directives; carries out this mission by managing the
Federal Information Processing Standards Program, developing Federal ADP standards
guidelines, and managing Federal participation in ADP voluntary standardization activities;
provides scientific and technological advisory services and assistance to Federal Agencies; and
provides the technical foundation for computer-related policies of the Federal Government.
The Institute consists of the following divisions:
Systems and Software — Computer Systems Engineering — Information Technology.
'Headquarters and Laboratories at Gaithersburg, Maryland, unless otherwise noted;
mailing address Washington, D.C. 20234.
^Some divisions within the center are located at Boulder, Colorado, 80303.
The National Bureau of Standards was ceorganized, effective Aprii 9, 1978.
Ultrasonic Tissue Characterization li
1 - .
A collection of reviewed papers based on talks presented at the *Q O OC)
Second International Symposium on Ultrasonic Tissue Characterization - ^
held at the National Bureau of Standards, Gaithersburg, Maryland U - 1
June 13-15, 1977 ^ ^
Edited by , .
Melvin Linzer
National Measurement Laboratory
National Bureau of Standards
Washington, DC 20234
Cosponsors of Symposium on Ultrasonic Tissue Characterization:
National Bureau of Standards
(National Measurement Laboratory)
National Science Foundation
(Research Applied to National Needs, RANN)
National Institutes of Health
(Diagnostic Radiology Department, Clinical Center)
U.S. DEPARTMENT OF COMMERCE, Juanita M. Kreos, Secretary
Jordan J. Baruch, Assistant Secretary for Science and Teclinology
NATIONAL BUREAU OF STANDARDS, Ernest Ambler, Director
Issued April 1 979
Library of Congress Catalog Card Number: 79-600026
National Bureau of Standards Special Publication 525
Nat. Bur, Stand. (U.S.), Spec. Publ. 525, 339 pages (Apr. 1979)
CODEN: XNBSAV
U.S. GOVERNMENT PRINTING OFFICE
WASHINGTON: 1979
For sale by the Superintendent of Documents, U.S. Government Printing Office, Washington, D.C. 20402
Stock No. 003-003-02058-3 Price $5.50
(Add 25 percent additional for other than U.S. mailing)
FOREWORD
The Second International Symposium on Ultrasonic Tissue Characterization
set the stage for establishing this series of meetings as the world's leading
forums for the dissemination of the most recent and advanced research in the
field. Undoubtedly, the Symposium contributed significantly to the contemporary
improvement in medical diagnosis. This improvement has resulted from the appli-
cation to instrumentation design of the results of measurements of the inter-
actions of ultrasound with tissue and from the deeper understanding of the
physical principles underlying these interactions.
The interdicipl inary approach of engineers, physical scientists, physicians,
and mathematicians made the Symposium a unique and fertile occasion. This
collection of reviewed papers which describes the research results presented at
the meeting makes it possible for the benefits of this work to be shared by the
whole community of those dedicated to the advancement of biomedical ultrasonics
and its application to health care.
The National Bureau of Standards is pleased to be responsible for this
publication and to have joined the National Science Foundation and the National
Institutes of Health as cosponsors of the Symposium.
John D. Hoffman
Director
National Measurement Laboratory
National Bureau of Standards
V
PREFACE
This volume forms a comparison with Ultrasonic Tissue Characterization, NBS Special
Publication 453 (1976). It contains extended versions of 43 of the 54 papers presented at
the Second International Symposium on Ultrasonic Tissue Characterization which was held at
the National Bureau of Standards on June 13-15, 1977. In a departure from the normal
practice for conference proceedings, these papers were critically reviewed by experts in
the field.
In the pages preceding the scientific papers, an overview of the. meeting is presented.
The first article in this chapter is a resume of the proceedings of the entire Symposium and
it serves to put the presentations and discussions into perspective. This is folowed by a
report summarizing the Panel Discussion on Breast Cancer, which was one of the highlights of
the meeting, and finally, by the introductory talk given by Dr. A. J. Eggers, Jr., who at
the time of the meeting was Assistant Director for Research Application, National Science
Foundation. The scientific papers presented at the meeting are grouped into chapters
devoted to: attenuation and velocity I: mechanisms; attenuation and velocity II: methodo-
logy and measurements; scattering and attenuation; scattering; tumor Doppler signatures;
propagation through bone and skull; image reconstruction; signal processing and pattern
recognition; and tissue viability and tissue phantoms. A survey of velocity and attenuation
data in mammalian tissue is included in an appendix.
The success of the Symposium was due to the dedicated efforts of many individuals.
Acknowledgment is given to the members of the Program committee (F. Dunn, A. C. Kak,
J. F. Greenleaf, J. G. Miller, R. C. Waag and P. N. T. Wells) and to the other individuals
who served as reviewers for this volume. Special appreciation is due to Peter Wells for
his contributions as cochai rperson of the Panel Discussion on Breast Cancer and for his
valuable collaboration in preparing the summaries of the Symposium and of the Panel
Discussion. The efforts of Ronald B. Johnson and Sara Torrence and their conference staffs
in the organization and management of the Symposium are gratefully acknowledged. Special
thanks is given to Rosemary S. Maddock who has provided the coordination and the editorial
assistance in the many phases of the preparation of this volume.
Melvin Linzer
Editor and
Symposium Chairperson
vi
ABSTRACT
The Second International Symposium on Ultrasonic Imaging and Tissue Charac-
terization was held at the National Bureau of Standards on June 13-15, 1977. The
meeting was cosponsored by the National Bureau of Standards, the National Science
Foundation, and the National Institutes of Health. This volume contains extended
and reviewed papers based on 43 of the 54 talks presented at the Symposium. Topics
covered include techniques for measurement of ultrasonic tissue parameters, the
dependence of tissue properties on physical and biological variables (e.g., ultra-
sonic frequency, temperature), mechanisms of ultrasonic tissue interactions, propa-
gation through bone and skull, tumor Doppler signatures, computerized tomography,
signal processing and pattern recognition, and tissue phantoms. A survey of velocity
and attenuation data in mammalian tissue is included in an appendix.
Key words: Absorption; attenuation; computerized tomography; Doppler;
impedance; medical diagnosis; microscopy; pattern recognition;
scattering; signal processing; tissue characterization; tissue
parameters; ultrasound; velocity.
In order to describe experiments adequately, it has been necessary to identify commercial
materials and equipment in this book. In no case does such identification imply recommendation
or endorsement by the National Bureau of Standards, nor does it imply that the material or
equipment is necessarily the best available for the purpose.
vi i
/
J
CONTENTS
Page
Chapter 1. OVERVIEW
Report on the Symposium 3
Report on Panel Discussion: Ultrasonic Diagnosis of Breast Cancer 11
Introductory Address 15
Chapter 2. ATTENUATION AND VELOCITY I: MECHANISMS
Elements of Tissue Characterization. Part I. Ultrasonic Propagation J
Properties 19
Johnston, R. L., Goss, S. A., Maynard, V., Brady, J. K. ,
Frizzell, L. A., O'Brien, W. D., Jr., and Dunn, F.
Absorption of Sound in Tissues ^ . . 29
Carstensen, E. L.
Mechanisms of Ultrasonic Attenuation in Soft Tissue 37
O'Donnell , M. and Miller, J. G.
Chapter 3. ATTENUATION AND VELOCITY II: METHODOLOGY AND MEASUREMENTS
Elements of Tissue Characterization. Part II. Ultrasonic Propagation
Parameter Measurements 43
Goss, S. A., Johnston, R. L., Maynard, V., Nider, L.,
Frizzell, L. A., O'Brien, W. D., Jr., and Dunn, F.
A Device for Measuring Ultrasonic Propagation Velocity in Tissue . . . . . 53
Sollish, B. D.
Measurement of the Temperature Dependence of the Velocity of Ultra- \y
sound in Soft Tissues 57
Bowen , T., Connor, W. G., Nasoni , R. L., Pifer, A. E., and
Sholes, R. R.
Ultrasonic Attenuation in Normal and Ischemic Myocardium 63
O'Donnell, M., Mimbs, J. W., Sobel, B. E., and Miller, J. G.
Acoustic Microscopic Analysis of Myocardium 73
Yuhas, D. E. and Kessler, L. W.
Acoustic Properties of Normal and Abnormal Human Brain 81
Kremkau, F. W., McGraw, C. P., and Barnes, R. W.
Frequency Dependent Attenuation of Malignant Breast Tumors Studied
by the Fast Fourier Transform Technique 85
Fry, E. K., Sanghvi , N. T., Fry, F. J., and Gallager, H. S.
Correlation of Ultrasonic Attenuation with Connective Tissue Content
in Breast Cancers 93
Kobayashi , T.
The Attenuation of Selected Soft Tissue as a Function of Frequency 101
Le Croissette, D. H., Heyser, R. C, Gammell, P. M.,
Roseboro, J. A., and Wilson, R. L.
Chapter 4. SCATTERING AND ATTENUATION
Clinical Spectrum Analysis Techniques for Tissue Characterization Ill
Lizzi, F. L. and Elbaum, M. E.
ix
Tissue Characterization In Vivo by Differential Attenuation
Measurements 121
L^vi, S. and Keuwez, J.
Statistical Estimation of the Acoustic Attenuation Coefficient
Slope for Liver Tissue from Reflected Ultrasonic Signals 125
Kuc, R., Schwartz, M., Finby, N., and Dain, F.
Chapter 5. SCATTERING
An Ultrasonic Tissue Signature for the Lung Surface 135
Rhyne, T. L.
Angle Scan and Frequency-Swept Ultrasonic Scattering Characteriza-
tion of Tissue 143
Waag, R. C, Lee, P. P. K., Lerner, R. M., Hunter, L. P.,
Gramiak, R., and Schenk, E. A.
Quantitative Measurements of Scattering of Ultrasound by Heart
and Liver 153
Reid, J. M. and Shung, K. K.
Dependence of Ultrasound Backscatter from Human Liver Tissue on
Frequency and Protein/Lipid Composition 157
Freese, M. and Lyons, E. A.
Ultrasound Backscattering from Blood: Hematocrit and Erythrocyte
Aggregation Dependence 165
Hanss, M. and Boynard, M.
Chapter 6. TUMOR DOPPLER SIGNATURES
Tumour Detection by Ultrasonic Doppler Blood-Flow Signals 173
Wells, P. N. T., Halliwell, M., Mountford, R. A.,
Skidmore, R., Webb, A. J'., and Woodcock, J. P.
Chapter 7. PROPAGATION THROUGH BONE AND SKULL
A Theory Relating Sonic Velocity to Mineral Content in Bone 179
Lees, S. and Davidson, C. L.
Ultrasonic Properties and Microtexture of Human Cortical Bone 189
Yoon, H. S. , and Katz, J. L.
Attenuation and Dispersion of Ultrasound in Cancellous Bone 197
Barger, J. E.
Transkull Transmission of Axisymmetric Focused Ultrasonic Beams
in the 0.5 to 1 MHz Frequency Range: Implications for Brain Tissue
Visualization, Interrogation, and Therapy 203
Fry, F. J.
Some Advances in Acoustic Imaging Through Skull 209
Smith, S. W., Phillips, D. J., von Ramm, 0. T. , and Thurstone, F. L.
Chapter 8. IMAGE RECONSTRUCTION
Characterization of In Vivo Breast Tissue by Ultrasonic
Time-of-Fl ight Computed Tomography 221
Glover, G. H.
Variation of Acoustic Speed with Temperature in Various Excised Human
Tissues Studied by Ultrasound Computerized Tomography 227
Rajagopalan, B., Greenleaf, J. F., Thomas, P. J., Johnson, S. A.,
and Bahn, R. C.
X
High Spatial Resolution Ultrasonic Measurement Techniques for
Characterization of Static and Moving Tissues 235
Johnson, S. A., Greenleaf, J. F., Rajagopalan, B., Bahn, R. C,
Baxter, B. and Christensen, D.
Mapping True Ultrasonic Backscatter and Attenuation Distribution
in Tissue - A Digital Reconstruction Approach 247
Duck, F. A. and Hill, C. R.
Chapter 9. SIGNAL PROCESSING AND PATTERN RECOGNITION
A Comprehensive Ultrasonic Tissue Analysis System 255
Linzer, M., Parks, S. I., Norton, S. J., Higgins, F. P.,
Dietz, D. R., Shideler, R. W., Shawker, T. H., and
Doppman, J. L.
Theoretical Analysis of Instantaneous Power Spectra as Applied
to Spectra-Color Ultrasonography 261
Jennings, W. D., Holasek, E., and Purnell, E. W.
Identification of Tissue Parameters by Digital Processing of
Real-Time Ultrasonic Clinical Cardiac Data 267
Joynt, L., Boyle, D., Rakowski, H., Popp, R., and Beaver, W.
Dynamic Autocorrelation Analysis of A-Scans Iji Vivo 275
Gore, J. C, Leeman, S., Metreweli, C, Plessner, N. J.,
and Willson, K. ■
Computer Spectral Analysis of Ultrasonic A Mode Echoes 281
Robinson, D. E.
Cepstral Signal Processing for Tissue Signature Analysis 287
Fraser, J., Kino, G. S., and Birnholz, J.
Recognition of Patterns in Ultrasonic Sectional Pictures of the
Prostate for Tumor Diagnosis 297
von Seelen, W., Gaca, A., Lock, E., Scheiding, W., and
Wessels, G.
Recent Developments in Obtaining Histopathological Information from
Ultrasound Tissue Signatures 303
Preston, K., Jr., Czerwinski, M. J., Skolnik, M. L., and
Leb, D. E.
Chapter 10. TISSUE VIABILITY AND TISSUE PHANTOMS
Damage and Death in Tissues and Associated Changes in Their
Mechanical Properties 317
Weiss, L.
A Human Abdominal Tissue Phantom 323
Edmonds, P. D., Reyes, Z., Parkinson, D. B., Filly, R. A.,
and Busey, H.
Tissue Simulators for Diagnostic Ultrasound 327
Eggleton, R. C. and Whitcomb, J. A.
Tissue Equivalent Test Objects for Comparison of Ultrasound
Transmission Tomography by Reconstruction with Pulse Echo
Ultrasound Imaging 337
Carson, P. L., Shabason, L., Dick, D. E., and dayman, W.
Appendi X
Data of the Velocity and Attenuation of Ultrasound in
Mammal ian Ti ssues - A Survey 343
Parry, R. J. and Chivers, R. C.
xi
I
Chapter 1
OVERVIEW
1
REPORT ON THE SYMPOSIUM
The Second International Symposium on Ultra-
', onic Tissue Characterization had six principal
objectives :
•to review the progress which had been made
J in the past two years since the first
' . Symposium
• to provide a forum for the exchange of ideas
among researchers, manufacturers, and
cl inicians
•to identify clinical problems which might be
solved by ultrasonic tissue characterization
I •to identify research opportunities
j •to promote the transfer of new technological
' advances in medical ultrasound to commercial
appl ication
' •to explore the potential for using ultrasonic
tissue characterization as a mass screening
! technique for breast cancer.
The Symposium, which extended over three days,
fas cosponsored by the National Bureau of Stan-
llards (NBS), the National Science Foundation (NSF),
'.nd the National Institutes of Health (NIH). The
introductory Session opened with a welcoming
iddress by Dr. E. Ambler, Director of NBS. He
'!poke of the rapid growth in the application of
jiltrasonic diagnosis. The U.S. market was esti-
mated to be $64M in 1977, increasing at 31 percent
i^er year. Ultrasonic diagnosis was apparently
;afe. Quantitative methods were being developed,
Accelerated by the work of the Tissue Signature
'roject supported by NSF. One of the most exciting
liossibilities was that ultrasound might be useful
iin mass screening for breast cancer. The interest
ijind involvement of NBS in these medical applica-
'|;ions was increasing. Dr. A. J. Eggers, Jr.,
'Assistant Director for Research Applications, NSF,
Emphasized the importance of interagency collabora-
ipion. He pointed out the dramatic decline in the
jcost of microelectronic systems; forecasts indica-
jced that tasks requiring microelectronics presently
:bosting $1M would cost ( 1 977 money) $100K in five
j/ears, and only $10K in ten years. He predicted
jthe imminent end of the "visible" computer age,
'jinticipating the development of displays with
J'ntegrated microelectronics for operator inter-
action. He referred to the 1973 findings of the
kudy group sponsored by NSF, and to the resulting
initiation of Experiment No. 5, which was designed
io test the acceleration of technology transfer to
■industry through the vehicle of diagnostic ultra-
50und. This had been followed by the preparation
py the Alliance for Engineering in Medicine and
Siology (AEMB), with NSF support, of a five-
/ear research and development agendum; 23 research
categories had been identified. Finally, in
addition to many project grants, NSF was supporting
the Tissue Signature Project of the Carnegie-Mellon
Institute of Research. Dr. Eggers was followed by
Dr. M. B. Lipsett, Director, Clinical Center, NIH.
Dr. Lipsett outlined the NIH program of support for
diagnostic ultrasonic research, presently running
at about $5M per year. The National Cancer Insti-
tute (NCI) was funding projects concerned with
breast pathology imaging and with endoscopy. The
National Heart and Lung Institute (NHLI) was
involved with real-time imaging and with Doppler
studies of blood flow. The National Institute of
General Medical Sciences (NI6MS) was supporting
research in imaging systems and the biological
effects of ultrasound. The Clinical Center was
developing a real-time scanner, and was collabora-
ting with NBS in the construction of a comprehensive
tissue analysis system. Dr. Lipsett stated that
NBS expected to continue to be concerned with the
advancing horizon of ultrasonic diagnostics.
The Symposium Chairperson, Dr. M. Linzer (NBS,
Gai thersburg , Maryland) discussed the present
knowledge of the ultrasonic properties of tissues.
He reviewed the advantages of ultrasonic diagnosis,
particularly in comparison with x-ray imaging. He
demonstrated these by state-of-the-art illustra-
tions of ultrasonic data obtained in several leading
laboratories -- Ultrasonics Institute, Sydney,
Australia, Duke University, Durham, North Carolina,
and Horizons Research Laboratories, Fort Lauderdale,
Florida. Comparably impressive results were begin-
ning to be obtainable with commercial instruments.
Dr. Linzer went on to discuss basic data necessary
for further development of ultrasonic diagnostics
and he spoke of new methods of display. He then
recalled the First International Symposium on
Ultrasonic Tissue Characterization which had been
held two years previously at NBS in Gai thersburg .
That meeting had been attended by more than 200
people and the proceedings were recorded in NBS
Special Publication 453. As a result of the
meeting, NSF had supported the Ultrasonic Tissue
Signature Working Group and the Advisory Council
of Users. Data on tissue parameters were being
updated, and libraries of algorithms and scans
were to be collected. In the present Symposium,
there were to be more than 50 papers, both from
within the United States and by 15 speakers from
nine countries outside the United States. It was
evident that measurements were more difficult to
make in the clinic than in the laboratory. Dr.
Linzer outlined the conference program which was
designed to focus sharply on measurement of tissue
characteristics. He emphasized that other consid-
erations were also vital for clinical success, and
that consequently the program for the Third Sympo-
sium, which would be held in June 1978, would have
an expanded scope and would include imaging and
Doppler techniques.
The first scientific session. Velocity and
Attenuation I: Mechanisms, began with R. L. Johnston,
S. A. Goss, V. Maynard, J. K. Brady, L. A. Frizzell,
W. D. O'Brien, Jr., and F. Dunn (University of
Illinois, Urbana, Illinois) reviewing ultrasonic
3
propagation properties. Different materials were
considered in order of increasing complexity. In
water, attenuation was proportional to the square
of the frequency, although the absolute value of
attenuation was rather greater than that which
would be expected to be due to viscosity. Solu-
tions of amino acids had similar properties to
those of water; while those of polypeptides
exhibited greater absorption probably due to such
interactions as helix-coil rearrangements and pro-
ton transfers. Proteins were made up of large
molecules, with relaxational absorption. Tissues
had characteristic attenuation dependency on
temperature and frequency. In general, both
increasing attenuation and velocity seemed to be
associated with decreasing water content and
with increasing protein and collagen content. Bone
and lung exhibited more complicated relationships.
For non-gas-containing soft tissues, however,
tissues with similar physiological functions had
similar attenuation, and this property might be
used to characterize tissues. In discussion, the
authors stated that the strong frequency dependence
of velocity in lung might be due to changing ultra-
sonic pathways, and they agreed that the role of
collagen deserved further study. E. A. Carstensen
(University of Rochester, Rochester, New York) then
reviewed the absorption of ultrasound in tissue.
He considered first the relatively small contribu-
tion to total absorption due to the cellular struc-
tural components of tissue. The linear frequency
dependence of absorption suggested a spectrum of
relaxation processes. The temperature coefficient
of attenuation had been measured for few tissues,
and data on dispersion were sparse. Experimental
results, however, were consistent with the theory.
Mechanisms of ultrasonic attenuation in soft tissue
were further discussed by M. O'Donnell and J. G.
Miller (Washington University, St. Louis, Missouri).
They presented a theoretical analysis of losses due
to cellular motion, and showed that in the limit
for small scatterers there was a linear frequency
dependence of attenuation. They concluded that
losses due to microscopic inhomogeneities accounted
for a substantial fraction of the total observed
attenuation in soft tissues. The authors were
questioned about the effects of concentration,
which they said they had taken into account, and
about the existence of dispersion, which they
believed to be due to relaxation.
Session 2, Scattering and Attenuation, dealt
with techniques for estimating attenuation from
measurements of scattering. F. L. Lizzi and M.
Elbaum (Riverside Research Institute, New York,
New York) described clinical spectrum analysis
techniques for tissue characterization. They had
measured the echo amplitude decrement from within
a narrow gated sample tracked increasingly deeply
into homogeneous tissue layers. Moreover, spec-
tral analysis revealed characteristic periodici-
ties for different histologies. Within the eye
and orbit, retina, hemorrhage, fat, melanoma and
glioma could be distinguished. Using a model
embodying scattering (from a fixed interface) and
attenuation, S. Levi and J. Keuwez (Hopital
Uni versitaire Brugmann, Brussels, Belgium) re-
ported an attempt to characterize tissues in vivo
by differential attenuation measurements on the
basis of observations made at two frequencies
(2 and 4 MHz). Results obtained in vivo with
leiomyoma and cyst were encouraging^ RT Kuc and
M. Schwartz (Columbia University, New York, New
York) and N. Finby and F. Dain (St. Luke's Hospital
Center, New York, New York), in a joint paper on
statistical estimation of the acoustic attenuation
coefficient slope for liver tissue from reflected
ultrasonic signals, discussed the separation of
fixed and refrigerated liver tissue on the basis
of spectral differences in the backscattered signals.
The liver was treated as a random linear filter.
This Session was closely related to Session 3,
Scattering. The surface of the lung was considered
by T. L. Rhyne (Massachusetts Institute of Tech-
nology, Cambridge, Massachusetts) to have an ultra-
sonic tissue signature which could be represented
by a stochastic model. Dr. Rhyne took special
care to measure the characteristics of the ultra-
sonic transmitter, transducer, and receiver which
he used. Recent developments in research into
angle-scan and frequency-swept ultrasonic scatter-
ing characterization of tissue were reported by
R. C. Waag, P. P. K. Lee, R. M. Lerner, L. P.
Hunter, R. Gramiak, and E. A. Schenk (University
of Rochester, Rochester, New York). Using an
experimental arrangement giving a true analog of
Bragg scattering, they demonstrated distinctive
correlations for normal, cirrhotic, and fatty
liver. Measurements of tissue in vitro were
related to models and measurements of known suspen-
sions. Some of the difficulties of measuring
scattering had been solved by F. E. Barber (Harvard
Medical School, Boston, Massachusetts), who
described ultrasonic microprobe methods for tissue
characterization. Dr. Barber's microprobe operated
at 10 MHz, and had an almost Gaussian focus with
a diameter of about 0.2 mm. Where large inter-
faces existed, they dominated the scattering. The
relationships between interface dimensions and
scattering directivity were elegantly demonstrated
by reference to a micrograph of the experimental
tissue. J. M. Reid and K. K. Shung (Providence
Medical Center and Institute of Applied Physiology
and Medicine, Seattle, Washington) presented quan-
titative measurements of scattering by heart and
liver. They had plotted the scattering cross-
sections of 12-20 cm' samples of tissue for many
orientations, and they discussed these in relation
to scattering equations for isotropic, plane, and
lossy scatterers. In particular, they considered
the effects of attenuation, both within the sample
and in the region between the transducer and the
sample. The dependence of ultrasound backscatter
from human liver tissue on frequency and protein/
lipid composition was discussed by M. Freese and
E. A. Lyons (Radionics Limited, Montreal, Quebec,
Canada). They had used a system with a bandwidth
equal to 25 percent of the center frequency, and
found that a frequency of 2.25 MHz gave the best
separation between normal and fatty liver. Signi-
ficant correlations of the backscatter with protein
content were observed in normal liver and with both
protein and lipid content in abnormal fatty liver.
Final lyjn this Session, M. Hanss and M. Boynard,
(Faculte de Medecine et de Biologie, Bobigny,
France) reported data on ultrasound backscattering
from blood and its hematocrit and erythrocyte
aggregation dependence. They stated that fluctua-
tions in backscattered amptitude were increased in
frequency with higher sedimentation rates during
the sedimentation process, and explained this by
a theory involving the existence of rouleaux in
Kigh sedimentation rate blood, and the occupancy
4
of potential scattering sites. They suggested
that eventually a noninvasive ultrasonic measure-
ment might supersede the contemporary pathology
test.
Session 4, Tumor Doppler Signatures, consis-
ted of two papers describing measurements of
blood flow changes due to malignancy. P. N. T.
Wells, M. Halliwell, R. A. Montford, R. Skidmore,
A. J. Webb, and J. P. Woodcock, (Bristol General
Hospital and Bristol Royal Infirmary, Bristol,
United Kingdom) explained the hypothesis that an
in situ carcinoma only becomes a rapidly prolif-
erating tumor after initiation of vascularization.
They showed that asymmetrical arterial blood flow
patterns might be found with two breasts, one
containing a malignant tumor. This was confirmed
in the following paper on Doppler echography by
G. Dale, Ch. M. Gros, M. Gautherie, and B. Gairard
(Senolgie-Hospices Civils, Strasbourg, France).
These authors had used a pulsed 8 MHz Doppler
system to study arterial flow patterns, and they
explained the complementary nature of thermography
in imaging venous flow. Returning to the paper by
Wells and his colleagues, these authors had gone
on to describe the discovery with an 8 MHz contin-
uous wave system of abnormal flow signals, appar-
ently associated with malignant tumor neovascular-
ization, arising from within the breast lesions
themselves. Such signals are not found in normal
breasts, or associated with benign tumors. They
reported that they had also detected similarly
augmented blood flow signals, using a 2 MHz pulsed
Doppler instrument, from within a malignant pan-
creatic tumor. In discussion, it was explained
that the breast tumor signals were sometimes
strongest at the edge of the lesion, perhaps
because of necrosis at the center.
Three papers reporting studies of cardiac
muscle were presented in Session 5, Attenuation
and Velocity II: Myocardium. T. D. Franklin, Jr.,
N. T. Sanghvi , F. J. Fry, K. M. Egenes, and A. E.
Weyman (Indiana School of Medicine and Indianapolis
Center for Advanced Research, Indianapolis, Indiana)
described ultrasonic tissue characterization studies
of ischemic and infarcted myocardium. The impetus
for this research came from the non-specificity
of conventional echocardiography in evaluating
infarction. Their preliminary results showed that
attenuation fell immediately after infarct. A more
extensive study by J. G. Miller, M. O'Donnell,
J. W. Mimbs, and B. E. Sobel (Washington University,
St. Louis, Missouri) of ultrasonic attenuation in
normal and ischemic myocardium, involved measure-
ments made with a cadmium sulfide phase-insensi-
tive receiving transducer. The range of attenua-
tion variation in normal myocardium was about 12
to 15 percent, and attenuation fell by about 20
percent when the temperature of the sample was
increased from 20° to 37 °C. The slope of a
least-squares line fitted to the attenuation
coefficient versus frequency data served as a con-
venient ultrasonic index. In dogs with experi-
mentally-induced ischemia, this slope was less
than the normal value if sacrifice was within
about 1 day after occlusion, but with later
killing it became greater. This may reflect an
increase in collagen content in necrotic scar
tissue. Acoustic microscopic analysis of myocar-
dium by D. E. Yuhas and L. W. Kessler (Sonoscan
Inc., Bensenville, Illinois) had yielded 100 MHz
data on attenuation and velocity in formalin-
fixed tissue. In kidney, it had previously been
shown that the effect on attenuation of formalin
fixing may not be significant. In normal myocar-
dium, the attenuation was about 3.5 to 5.7 dB cm"^
MHz-^. Moreover, the detailed demonstration at
high frequency of structures which were of wave-
length-order dimensions at low megahertz
frequencies was directly relevant to the study
of scattering in conventional echography.
The second day of the Symposium began with
Session 6, Image Reconstruction. G. H. Glover
(General Electric Co., Milwaukee, Wisconsin)
discussed in vivo measurement of ultrasonic refrac-
tive index distributions in human breasts by time-
of-f light tomography. His laboratory instrument
was a 5 MHz transmission computerized tomograph.
The patient lay prone on a canvas sling with one of
her breasts immersed in water at 32° to 36 °C.
Nine minutes were required to scan a slice of 5 mm
thickness. Reconstructed two-dimensional images
revealed cysts and fibroadenomas (having velocities
around 2 percent higher than water) and malignant
tumors (having velocities about 4 to 5.5 percent
lower). Young normals (age range 24 to 29 years)
had quite wide velocity variations, but they
might not need to be scanned in a mass screening
program. Histograms of pixel velocity distribu-
tions, and cumulative pixel velocity plots, might
provide accurate separation of normals from
patients with breasts made asymmetrical by the
presence of malignancy. Dr. Glover stated that he
was developing a fan beam detector array with 127
elements, designed to reduce the examination time
to 10 seconds per breast. Questions were raised
by the audience concerning path straightness (Dr.
Glover had ray plots), resolution (which was 1 ns,
although 10 ns would have been adequate), point
spread function (MTF - 1 mm), and accessibility
(slices could be within 20 mm of the chest wall
in the experimental system). Using a synthetic
"fan beam," R. Bal asubramani an , J. F. Greenleaf,
P. J. Thomas, and S. A. Johnson (Mayo Clinic,
Rochester, Minnesota) reported measurements of
temperature coefficients of ultrasonic speed in
various human tissue. Apart from that of fat, the
temperature coefficients of liver, kidney, and
other tissues were all about 2 m s"^K"^. The
possibility that tumors might be identifiable
because of their elevated temperatures was men-
tioned during the discussion, but the authors had
not investigated velocity changes with temperature
in tumor tissue. Results obtained with computed
tomographic reconstruction of attenuation images
had previously been disappointing, but new
approaches were suggested by A. C. Kak and K. A.
Dines (Purdue University, West Lafayette, Indiana)
in their paper on signal processing for the
measurement of attenuation of soft tissue. They
pointed out that reflection contributed a loss of
about 4 percent at each major interface, making
algebraic reconstruction techniques unsatisfactory.
Other frequency and time domain methods, such as
polynomial fitting, frequency averaging and
energy analysis, might prove to be good candidates
for attenuation measurements. For example, a
reconstruction of a section through a dog's
heart by the energy method demonstrated immunity
to reflection artifacts. In discussion, members
of the audience emphasized the importance of
clinical practicability. S. A. Johnson, J. F.
5
Greenleaf, and B. Rajagopalan (Mayo Clinic,
Rochester, Minnesota) and R. C. Bahn and B.
Baxter (University of Utah, Salt Lake City, Utah),
in a joint paper on the future role of high
spatial resolution ultrasonic measurement tech-
niques for characterization of static and moving
tissue, discussed the determination by computer-
ized tomography of three-dimensional fluid flow
and temperature. They proposed the use of
seismological approaches for high-resolution
imaging. Another new method, possibly leading
to mapping true ultrasonic backscatter and
attenuation distributions in tissue -- a digital
reconstruction approach -- was suggested by F. A.
Duck and C. R. Hill (Institute of Cancer Research,
Sutton, United Kingdom). By assuming that
attenuation and scattering were constant and iso-
tropic within each volume element of the object,
they showed that the corresponding pixel values in
the image could be calculated by algebraic recon-
struction techniques (ART) in about six iterations.
Discussion revealed that isotropic assumptions
might not be satisfied in clinical practice.
Sessions 7 and 8, Signal Processing and Pattern
Recognition I and II, contained a total of twelve
papers. Region-of-interest analysis methods were
used in feature extraction techniques designed to
improve recognition of patterns in ultrasonic
pictures of the prostate for tumor diagnosis, by
W. V. Seelen (Johannes Gutenberg Universitat,
Mainz), E. G. Lock and G. Wessels (Deutsche Klinik
fiir Diagnostik, Wiesbaden), U. Scheiding (Battelle
Institut, Frankfurt), and A. Gaca (Deutsche Klinik
fiir Diagnostik, Wiesbaden, Federal Republic of
Germany). J. C. Birnholz (Harvard University,
Boston, Massachusetts) then reviewed the elements
of visual pattern recognition in ultrasonography.
He showed some remarkably excellent grey-scale
pictures. The influence of signal processing, and
particularly of the detection method, on the quality
of echograms was discussed by I. Beretsky, D. Arnold,
and J. Cason (Searle Ultrasound Research and Ad-
vanced Development, Spring Valley, New York), in
their paper on impulse detection in pulse echo
ultrasound -- recent in vitro experiments with a
human aorta. This was followed by a theoretical
analysis of instantaneous power spectra as applied
to spectra-color ultrasonography, by W. D. Jennings,
E. Holasek, and E. W. Purnell (Case Western Reserve
University, Cleveland, Ohio). The method was
based on the separate display in the primary colors
on the same image of three frequency bands consti-
tuting the echo signals. Session 7 ended with a
paper by C. K. Kuni (University of Colorado
Medical Center, Denver, Colorado) on tissue iden-
tification by spectral analysis of scattered ultra-
sound. Dr. Kuni discussed some of the problems of
obtaining and interpreting backscattered spectra.
The following Session 8 began with a descrip-
tion of a comprehensive ultrasonic tissue analysis
system by M. Linzer, S. I. Parks, R. W. Shideler,
S. J. Norton, F. P. Higgins, and D. R. Dietz
(National Bureau of Standards, Washington, D.C.)
and J. L. Doppman and T. H. Shawker (National
Institutes of Health, Bethesda, Maryland). Dr.
Linzer described a dynamically-focused system
using an expanding-aperture annular array with an
approximately constant F-number, and an rf sub-
system capable of operating with impulses, gated
continuous wave, and chirp pulses. Chirp operation
with 8:1 compression had been achieved. An
ultrafast signal averager was developed for im-
provement of the signal-to-noise ratio of A-scans.
This instrument had a sampling rate of 50 MHz and
a 4-bit analog-to-digital converter, and had
potential applications in the examination of obese
patients, the use of higher frequencies, and the
use of inefficient transducers. Next, an instru-
ment called the "SonoChromascope" was described.
Interfaced to a commercial two-dimensional B-
scanner, this device formed a versatile digital
real-time acquisition, signal processing, and
display system. The 250 kbyte memory allowed
operation in normalized averaging, maximum,
minimum, and parameter comparison modes. Gray-scale,
color, and window displays, as well as digital
readout of area and average echo amplitude in
regions-of-interest were possible. Finally,
Dr. Linzer stated that other methods, including CT
scanning, were being investigated at NBS. There
was discussion about the rather low digitization
accuracy of commercial transient recorders which
do not contain sample-and-hold circuits. The
subject of spectrum analysis was again considered
in the following paper by L. Joynt, D. Boyle, H.
Rakowski, and W. Beaver (Stanford University,
Stanford, California), on the identification of
tissue parameters by digital processing of real-
time ultrasonic clinical data. Data obtained with
a real-time phased array sector scanner from
selected regions within the myocardium in various
clinical conditions were subjected to spectral
analysis. Although there was substantial overlap
between groups, spectra from patients with myocar-
dial infarcts had much less fluctuation in their
means and variances as a function of time. J. C.
Gore and S. Leeman (Royal Postgraduate Medical
School, London, United Kingdom) presented two short
papers. The first, on the theoretical evaluation
of backscatter of ultrasonic pulses from human
tissue and its implications for tissue characteri-
zation, dealt with the effects of the characteris-
tics of pulser, transducer, and receiver. The
second paper, on autocorrelation analysis of A-
scans in vivo and the possible clinical application
of temporal changes in echo characteristics,
revealed similarities between the echoes from
different specimens of the same types of tissue,
while different types of tissue had dissimilar
autocorrelation functions. With the goal of imple-
menting computer spectral analysis of ultrasonic
A-mode echoes, D. E. Robinson (Ultrasonics Institute,
Sydney, Australia) had developed an on-line data
acquisition system based on a clinical scanner,
10 MHz sample rate transient recorder, and Inter-
data 85 computer. Special consideration had been
given by K. Preston, Jr. (Carnegie-Mellon Univer-
sity, Pittsburgh, Pennsylvania) and M. L. Skolnik
(University of Pittsburgh, Pittsburgh, Pennsylvania)
to the application of pattern recognition techniques
to the separation of ultrasonic echoes from differ-
ent types of tissue. Digitized A-scans had been
converted to sound patterns as a preliminary to
testing the feasibility of using the ear to dis-
tinguish the sound patterns from different types
of tissue. The A-scans had been analyzed in terms
of skew, kurtosis, and other statistical parameters.
Initial results had been obtained from normal and
neoplastic kidney. In order to determine the
correlation function of echoes from a region-of-
interest identified on a two-dimensional B-scan,
J. Eraser and G. S. Kino (Stanford University,
6
Stanford, California) and J. Birnholz (Harvard
Medical School, Boston, Massachusetts) reported
the use of cepstrum processing for tissue analysis.
The transducer response appeared mainly centered
at the cepstrum origin, and it was gated out so
that the inverted cepstrum allowed the well-decon-
voluted autocorrelation response of the tissue
alone to be observed. Finally in this lengthy
pair of sessions, R. D. Lepper, R. Reuter, and
H. G. Trier (Universitat Bonn, Bonn, Federal
Republic of Germany) pointed out the advantage of
asychronous writing and reading on a storage tube
in time-stretching ophthalmic A-scans in order to
improve the precision of digitization.
Session 9 was a Panel Discussion on Ul tra-
sonic Diagnosis of Breast Cancer. The cochair-
person, P. N. T. Wells (Bristol General Hospital,
Bristol, United Kingdom) in his introductory remarks
pointed out that mortality from breast cancer had
not improved over the last 20 years. He explained
that in the Western World, breast cancer, which had
many different pathological types, killed 1 in 50
women, was the main cause of death in females in
the age range 40 to 44, and in the U.S. had an
annual economic cost of $200M. Existing diagnostic
methods were inadequate for mass screening. Interest
in ultrasonic techniques hinged around the possi-
bility that earlier detection by screening might
improve prognosis. Cochai rperson W. Pomerance
(National Institutes of Health, Bethesda, Maryland),
defined screening as the application of known
methods to an asymptomatic population to detect
early cancer. He considered that a successful
method for breast screening would need to detect
tumors of less than 5 mm diameter, and that it
should be free from hazard and have at least a 90
percent success rate. Encouraging results with
high-resolution scanning led L. Weiss (Roswell
Park Memorial Institute, Buffalo, New York) to
suggest that ultrasound might be capable of detec-
ting lesions of 2 mm diameter. This would be an
important although numerically modest improvement
over the present ability to detect 5 mm lesions by
manual palpation. He explained that it had been
estimated that screening the whole adult female
population with an effective ultrasonic method, if
one should be developed, would have cost $620M per
year in 1971, and he felt that it would therefore
be necessary to restrict screening to women known
to be at risk, or over 50 years old. J. L.
Doppman (National Institutes of Health, Bethesda,
Maryland) described the role of mammography, and
stated that the x-ray exposure was presently less
than 0.01 Gy. Mammography could ensure that the
lesion was removed by the surgeon. Five years of
experience of ultrasonic visualization of the
breast in Japan were reviewed by T. Kobayashi
(National Cancer Center Hospital, Tokyo, Japan).
Ultrasonic screening was being made available by
means of mobile units. G. Dale (Senologie-Hospices
Civils, Strasbourg, France) described the comple-
mentary use of ultrasonic scanning, both pulse-
echo and Doppler, together with thermography and
mammography, to decrease the overall error rate.
The results of research into high resolution pulse-
echo imaging of the breast, and the problems of
analyzing the scans, were presented by G. Baum
(Albert Einstein College of Medicine, New York,
New York). Elizabeth K. Fry (Indiana University
School of Medicine and Indiana University Hospital,
Indianapolis, Indiana) described a versatile ultra-
sonic breast scanning research instrument. She
presented some scans, and discussed their interpre-
tation, emphasized the importance of shadows, and
the different patterns obtained with different
transducers and frequencies. Cautious optimism
was expressed by D. E. Robinson (Ultrasonics
Institute, Sydney, Australia). He demonstrated
some excellent scans, obtained with a 4 MHz
focused beam in a water bath scanner, and spoke
of eight features for which the scans required to
be examined in reaching a diagnosis. J. F.
Greenleaf (Mayo Clinic, Rochester, Minnesota)
described the role of ultrasonic computed tomog-
raphy in obtaining data on tissue properties and
in correcting two-dimensional B-scans. G. H.
Glover (General Electric Company, Milwaukee,
Wisconsin) had previously talked about his work
on ultrasonic computed tomography in a paper pre-
sented during Session 6. As time was short. Dr.
Glover did not add to the information which he had
already given concerning the increase in velocity
associated with malignancy in breast tissue. For
the same reason. Dr. Wells did not elaborate on
the potential of neovascularization blood flow
Doppler detection, as he had mentioned this in
Session 4. M. Linzer (National Bureau of Standards,
Washington, D.C.) emphasized the vast quantity of
data which would be obtained in scanning only one
patient, and of the refinements in methodology
which would be necessary to make ultrasonic breast
screening feasible. Members of the Panel and the
audience then began an open discussion, and several
points were made. The demonstration of microcalci-
fication in ultrasonic scans was possible if
sufficiently extensive. Ultrasonic scanning was
beneficial in young women with "lumpy" breasts.
Mammography did not contribute further when ultra-
sound had already detected a lesion. Present
commercial equipment was unsuitable for starting a
mass screening program. Mammography would in
principle never be an acceptable mass screening
method, as it would always involve exposure to
some radiation. Dr. Wells closed the discussion
with a short summary. Mass screening of selected
groups of the female population would be worth-
while if a suitable technique could be developed.
Such a technique might well be based on ultrasound
but it would need to be accurate and economical.
The present lines of research were promising;
more resources should be devoted to solving the
fundamental and technological problems, so that
earlier clinical application would be possible.
Propagation Through Bone and Skull was the
subject of Session 10, which opened the final day
of the Symposium. A theory relating sonic velocity
to mineral content in bone was presented by S. Lees
(Forsyth Dental Center, Boston, Massachusetts).
Bone had a higher velocity than that predicted on
the basis of its longitudinal elastic modulus.
Dr. Lees suggested that this was due to the bone
consisting of mineral hydroxyapati te crystallites
embedded in a matrix of collagen, behaving in the
same way as a plastic filled with powdered mineral.
H. S. Yoon and J. L. Katz (Rensselaer Polytechnic
Institute, Troy, New York) considered the ultra-
sonic properties and microtexture of human cortical
bone, and showed that bone could be treated as an
elastic dielectric. The attenuation of ultrasound
in cancellous bone was the subject of a paper by
J. E. Barger (Bolt Beranek and Newman, Cambridge,
Massachusetts). He distinguished between phase
7
and group velocities, and showed that scattering
in the dipole made an important contribution to
losses at frequencies in the range 0.5 to 2 MHz.
The results of measurements of transmission through
skull were presented, in a discussion of the
acoustic characteristics of the skull, by D. N.
White (Queens University, Kingston, Ontario,
Canada). Dr. White explained that propagating
ultrasonic energy appeared to be distributed
between the bony and the soft tissue elements of
cancellous bone resulting in substantial
attenuation, although it was transmitted with
relatively low loss through ivory bone. Inhomo-
geneities led to marked distortion of megahertz
frequency beams transmitted through the skull.
Because of this difficulty, F. J. Fry (Indianapolis
Center for Advanced Research, Indianapolis, Indiana)
had investigated the transkull transmission of axi-
symmetric focused ultrasonic beams in the 0.5 to
1 MHz frequency range, and its implications for
brain tissue visualization, interrogation, and
therapy. Beams at the relatively low frequency of
0.5 MHz were not greatly distorted in travelling
through skull, and there seemed to be quite good
correlation between transkull ultrasonic two-
dimensional brain scans and x-ray computed tomo-
graphs. The potential for therapy resulting from
the relatively low attenuation in the skull was
demonstrated by a lesion induced in lucite by a
beam of ultrasound which had passed through part
of a cadaver skull. The feasibility of visualizing
the brain was further demonstrated by some advances
in acoustic imaging through the skull which were
reported in a joint paper by S. W. Smith (Bureau
of Radiological Health, Rockville, Maryland), D. J.
Phillips (University of Washington, Seattle, Wash-
ington), and 0. T. von Ramm and F. L. Thurstone
(Duke University, Durham, North Carolina). Remark-
able real-time pictures, showing brain structures
and pulsating arteries, had been made with a
phased-array sector scanner. Treating the skull
as a random acoustic lens, optimal results were
theoretically obtained at a frequency of about
1 MHz and an aperture width of about 35 to 40 mm.
The ultrasonic images were well correlated with
pictures from a brain atlas. The possibility was
pointed out that phase correction for the effect
of the skull could be obtained element-by-element
across the array, if a suitable intracranial source
of uniform wavefronts could be devised.
The following Sessions 11 and 12, Attenuation
and Velocity III and IV had ten papers. S. A. Goss,
R. L. Johnston, V. Maynard, L. Nider, L. A. Frizzell,
W. D. O'Brien, Jr., and F. Dunn (University of
Illinois, Urbana, Illinois) reviewed ultrasonic
propagation parameter measurements. They discussed
the advantages, disadvantages, and accuracies of
resonant cavities, interferometers, and thermo-
couple probes, and techniques based on velocity
difference measurement, pulse superposition, radia-
tion pressure, and time-of-f 1 ight measurement.
The temperature dependence of the velocity of sound
in soft tissues, a parameter of great importance in
hyperthermia therapy for control of cancer, was
reported by T. Bowen, W. G. Connor, R. L. Nasoni,
A. E. Pifer, and R. R. Sholes (University of
Arizona, Tucson, Arizona). With the exception of
fat, vegetable oil, and water, which had negative
temperature coefficients of velocity, all the other
tissues investigated -- spleen, muscle, liver,
and kidney -- had positive temperature coefficients
over the range 36° to 44 °C. A simple but effec-
tive device for measuring ultrasonic propagation
velocity in tissue, both in vitro and, with
suitable anatomy, in vivo, was described by B. D.
Sollish (Weismann Institute of Science, Rehovat,
Israel). The following three papers were con-
cerned with ultrasonic studies of the breast. T.
Kobayashi (National Cancer Center Hospital, Tokyo,
Japan) spoke about the correlation of attenuation
in breast cancers with connective tissue content.
He showed that stronger shadowing, associated
with greater attenuation, was correlated with
higher connective tissue content. Thus, strong
shadows were produced by scirrhous carcinoma
(rich in connective tissue), average shadows by
papillary carcinoma, and weak shadows by medullary
carcinoma (poor in connective tissue). The
comments of the audience substantiated the conclu-
sion that increasing attenuation was correlated
with increase in collagen, as in operative scars
and older breasts. Next, G. Dale, Ch. M. Gros,
M. Gautherie, and B. Gairard (Senologie-Hospices
Civils, Strasbourg, France) carefully reviewed
diagnostic image features, in their paper on echo-
graphic syndromes of breast cancer. This was
followed by a mul ti -di sci pi ine approach to the
detection of breast cancer by ultrasonic techniques,
with intercomparison of signal-processed ultra-
sound transmission data, ultrasound pulse-echo
information, and whole breast pathology, by E. K.
Fry, N. T. Sanghvi , and F. J. Fry (Indiana Univer-
sity School of Medicine and Indianapolis Center for
Advanced Research, Indianapolis, Indiana) and H. S.
Gallager (University of Texas, Houston, Texas).
Experiments with an excised, formalin-fixed breast
revealed non-uniform attenuation particularly in
and near the nipple. The audience asked about the
contributions of reflections (which were felt not
to be large), the possibility of the existence of
bubbles and the effect of formalin, and, for in
vivo measurements , the effects of lactation and
the taking of the contraceptive pill. F. W. Kremkau,
C. P. McGraw, and R. W. Barnes (Bowman Gray School
of Medicine, Winston-Salem, North Carolina) reported
the results of some careful measurements of the
acoustic properties of human brain. The infant
brain had about 60 percent lower attenuation than
the adult. In adult brain, fixing with formalin
increased velocity by about 30 percent. Fixed
brain had a dispersion of about 1.6 m s'^MHz"^, and
fresh brain about 2 m s"^MHz-i. The temperature
coefficient of attenuation was negative, and
attenuation was 1.4 times greater in white matter
than in gray. Tissue characterization using
acoustic transmission and scattering parameters was
further discussed in a joint paper by M. P. Kadaba,
W. P. Cockerill, and P. K. Bhagat (University of
Kentucky, Lexington, Kentucky), and R. W. Ware
(Veterans Administration Hospital, Lexington,
Kentucky). In tissues such as kidney, liver, and
cardiac muscle, over the frequency range to 1 to
10 MHz, attenuation increased from about 2 to 16 dB
cm-^, and velocity, from 1520 to 1580 m s"' (i.e.,
by about 4 percent). In another joint paper, D. H.
Le Croissette, R. C. Heyser, P. M. Gammell, and
J. A. Roseboro (California Institute of Technology,
Pasadena, California), and R. L. Wilson (University
of Southern California, Los Angeles, California)
reported values of the attenuation of selected
tissue as a function of frequency. Using time-delay
spectrometry, involving the measurement of the
frequency difference between the received signal
8
and a transmitted frequency-modulated wavetrain,
they studied attenuation over the range 1 to 8 MHz
in liver, pancreas, muscle, and fat. In liver,
for example, attenuation decreased after death but
increased as a result of fixing. Apparently
similar tissues often had very different attenua-
tions. Finally in Session 12, D. Hughes, L. A.
Geddes, and V. Newhouse (Purdue University, West
Lafayette, Indiana) discussed the velocity and
attenuation of ultrasound in blood at 37 °C. These
data were required in order to estimate Young's
modulus of the aortic wall. Besides presenting
accurate values, Hughes reported that attenua-
tion increased with packed cell volume (up to at
least 60 percent packing), and velocity had a
minimum value with a packed cell volume of about
10 percent.
The final session. Session 13, Tissue Viability
and Tissue Phantoms, began with a paper by L. Weiss
(Roswell Park Memorial Institute, Buffalo, New York)
on tissue signatures -- a matter of life and death.
Dr. Weiss discussed the changes which took place in
1n vitro tissues, and the precautions which needed
to be taken to minimize consequential changes in
tissue signature interactions. For example, poor
oxygenation might result in death within 5 minutes.
Even in life, many tumors had dead or dying cells,
and these differences probably affect ultrasonic
imaging. Mechanically-induced cell separation
measurements were related to degenerative changes
in tissue, and might possibly be used to standardize
interactions with ultrasound. The remaining three
papers in this Session were concerned with tissue
equivalent materials for ultrasonic imaging. In a
joint paper, P. Edmonds, Z. Reye, and D. Parkinson
(Stanford Research Institute, Menlo Park, Califor-
nia), R. Filly (University of California Medical
Center, San Francisco, California), and H. Busey
(Picker Corporation, Northford, Connecticut)
described a human tissue phantom for testing con-
ventional ultrasonic scanners. After experimenting
with several rubber and plastic materials, it was
found that gelatin-water gels, loaded with
glass microspheres or cellulose fillers as
scatterers, gave satisfactory results. Questions
from the audience were answered by statements that
the gel had an acceptable dependence of attenua-
tion on frequency, and that the plastic materials
had attenuations which were too high to permit
the addition of scatterers. R. C. Eggleton (Indiana
University School of Medicine and Indianapolis
Center for Advanced Research, Indianapolis, Indiana)
mentioned the application of tissue simulators for
ultrasonic diagnosis, in teaching and training,
evaluation of scanner performance, and as models
for basic research. He presented results obtained
with a plastisol, and showed the effects of adding
scatterers. A joint paper by P. L. Carson, L.
Shabason, and D. E. Dick (University of Colorado
Medical Center, Denver, Colorado) and W. dayman
(Alderson Research Laboratories, Inc., Denver,
Colorado) on tissue equivalent test objects for
comparison of ultrasound transmission tomography
by reconstruction and pulse echo imaging, also
discussed special plastic materials. Although
these materials did have rather high attenuation
(about 2 dB cm-iMHz"-), scatterers could be added
and phantoms satisfactory for testing systems
could be constructed. Finally, Dr. Carson showed
an ultrasonic CT scan of an in^ vi vo breast indica-
ting an attenuation coefficient of about 15 dB cm"^
at 3.5 MHz. •
In summary, the Second International Symposium
on Ultrasonic Tissue Characterization brought the
subject into focus and perspective. The 286 par-
ticipants, including many of the leaders in the
field, spent three days exchanging ideas and
learning from each other. Mass screening for
breast cancer in particular was identified as one
of several clinical problems which might be solved
by ultrasonic tissue characterization. Research
opportunities were evident in the development of
fundamental theories, the acquisition of data on
velocity, attenuation and scattering, and in
clinical validation. Image reconstruction, signal
processing, pattern recognition, and Doppler blood
flow signals, emerged as fruitful areas for inves-
tigation. New technological advances, for example,
in tissue phantom materials, were identified as
already being ready for transfer to commercial
application.
M. Linzer
P. N. T. Wells
July, 1977
9
REPORT ON PANEL DISCUSSION
ULTRASONIC DIAGNOSIS OF BREAST CANCER
PANEL
Cochairpersons
William Pomerance
Diagnosis Branch
Division of Cancer Biology and Diagnosis
National Institutes of Health
Bethesda, Maryland
P. N. T. Wells
Department of Medical Physics
Bristol General Hospital
Bristol, United Kingdom
Panel Members
Gilbert Baum
Albert Einstein College of Medicine
Bronx, New York
John L. Doppman
Department of Diagnostic Radiology
Clinical Center
National Institutes of Health
Bethesda, Maryland
G. H. Glover
General Electric Company
Medical Systems Division
Milwaukee, Wisconsin
Toshiji Kobayashi
Department of Internal Medicine
National Cancer Center Hospital
Tokyo, Japan
D. E. Robinson
Ultrasonics Institute
Sydney, Australia
G. J. Dale
Senologie-Hospices Civils
Strasbourg, France
Elizabeth K. Fry
Indiana University School of Medicine and
Indiana University Hospital
Indianapolis, Indiana
J. F. Greenleaf
Mayo Clinic
Rochester, Minnesota
Melvin Linzer
National Measurement Laboratory
National Bureau of Standards
Washington, D.C.
Leonard Weiss
Roswell Park Memorial Institute
Buffalo, New York
In his introductory remarks, P. N. T. Wells
pointed out that the overall mortality from breast
cancer was the main cause of death in Western women
between the ages of 40 and 44 years, and the annual
economic cost of the disease in the U.S. was around
$200M. Consequently, it was timely to review the
contribution which ultrasonic diagnostics might
make to the solution of the breast cancer problem.
There were many different malignant tumors of the
breast — scirrhous, mammary duct, papillary,
medullary, colloid, lobular, intracystic, apocrine
and adenoid cystic carcinoma, Paget' s disease,
lymphoma, and sarcoma. The earlier that breast
cancer was treated, the better was the prognosis.
A patient treated with in situ carcinoma was cured.
If the tumor was localized, the five-year survival
was 85 percent; but this fell to 53 percent if
there was lymph node involvement. A patient with
distant metastases was incurable, although the
survival time varied from patient to patient.
Interest in screening hinged around the possibili-
ty that earlier detection might improve prognosis.
Unfortunately, thermography was unsatisfactory as
a screening method for breast cancer: in one
series, for example, only 25 percent of those who
developed cancer within 18 months were detected,
and the false positive rata was approaching 15 per-
cent. When combined with clinical examination, in
another trial, only about 20 percent of cancers in
women under 50 years of age would not have been
detected without mammography. Moreover, mammogra-
phy, whilst having a negligible risk as far aS
the individual patient was concerned, would itself
induce by radiation a large number of cancers if
every member of a large population were to be
exposed to it.
William Pomerance said that it had to be
admitted that contemporary methods of treating
advanced breast cancer were ineffective, and that
progress might depend on earlier detection. He
defined screening as the application of known
methods to an asymptomatic population to detect
early cancer. He considered that an acceptable
method for breast cancer screening would need to
detect tumors of less than 5 mm diameter, and
that it should be free from hazard and have at
least a 90 percent success rate. The assessment
of the value of breast cancer screening would take
a long time, since the mortality of patients sur-
viving treatment only coincided with that of the
normal population after about 17 years. He thought
that screening might have its biggest impact in
detecting slowly growing asymptomatic tumors.
11
The fact that screening for cancer of the
uterine cervix had resulted in a reduction in
mortality led Leonard Weiss to hope that breast
cancer screening might also be effective. Moreover,
he was encouraged by results which he had obtained
using a high-resolution ultrasonic scanner, that
it might soon become possible to detect lesions
of less than 2 mm diameter, although it might be
difficult to distinguish histologically between
different types of lesion. The detection of 2 mm
diameter lesions would be an important but numeri-
cally modest advance from the present ability to
detect 5 mm lesions by manual palpation; an improve-
ment of 10 percent in effectively treated patients
might result. Weiss stressed the importance of
growth rate in determining prognosis. He then
referred to a study of the feasibility of breast
cancer screening, which had been carried out in
1971. If the 38 million women in the U.S. who
might have benefited from screening were to have
been given this advantage, it had been estimated
that the annual cost then would have been around
$620M. He felt that it would therefore be necessary
to restrict screening to women known to be at risk,
or over 50 years old.
The role of mammography was discussed by J . L.
Doppman. He showed some typical mammograms, and
explained that the method was best for visualizing
microcalcif ication. Three-dimensional display was
needed. A great advantage of mammography was that
it could be used by the surgeon to ensure that he
removed the entire lesion. Recent developments had
resulted in a reduction in the x-ray exposure,
which presently was generally less than 0.01 Gy
(1 rad).
Next, experience with pulse-echo ultrasonic
methods was reviewed. Toshiji Kobayashi had worked
for five years in Japan on the ultrasonic visuali-
zation of the breast. For Japanese women, scanning
through a flexible membrane at the bottom of a
water-bath was satisfactory. Kobayashi described
the distinctive echographic features of lesions
of differing histologies, and he emphasized the
importance of retrotumorous shadowing in deducing
information about the attenuation by the lesion.
Such was the confidence which was felt in the
method in Japan that specially equipped minibuses
were being used in mass screening trials. In
the breast cancer clinic, both pulse-echo and
Doppler investigations were complementary to
thermography and mammography, according to G. J.
Dale. He discussed the ultrasonic scans of two
typical patients, and concluded that in breast
cancer diagnosis the false positive rate was 0.6
percent; the false negative rate was 11 percent,
but this fell to 5 percent when combined with
clinical examination. Using pulsed Doppler, he
had traced breast arteries of greater than 1 to
2 mm in diameter, and he explained that Doppler
information related more to function than to
structure. He had not studied the capillary
system of malignant tumors by the Doppler method.
Gilbert Baum described the interpretation of
breast echograms. Not only was textural analysis
of breast tissue patterns difficult because of
their wide variations, but also the skin of
different individuals seemed often to transmit
ultrasound differently. He felt that it was neces-
sary to scan through water, particularly because
lesions might otherwise be displaced by the pressure
of the ultrasonic probe. Coupling through a water
bath also allowed a large aperture transducer to be
used, and the scanning motion could be automatic.
Apart from the need to develop instrumentation
capable of yielding reproducible results, the main
problem was not to obtain the scans, but to find
the time to analyze them. Furthermore, it was sub-
jectively apparent that image degradation occurred
when digitized either to fewer than 360 x 360
pixels, or to less than 32 gray levels. Sixty
scans were obtained for each patient, and to reduce
the analysis time he had tried color-coding, and
the superimposition of seven serial scans on a
three-dimensional optical hologram. Even with such
techniques, it was obvious that screening could
only be offered to selected groups of patients.
Elizabeth K. Fry had constructed a versatile
scanner for research into breast disease. This
was important, because the modification of instru-
ments primarily designed for the examination of
other structures was unsatisfactory. The patient
lay prone, her breasts immersed in water. A
variety of transducers, operating at different
frequencies, had been used to make many scans,
producing a great range of patterns for analytical
study. Cautious optimism was expressed by D. E.
Robinson because of his experience with a water-
immersion breast scanner using a 4 MHz transducer,
with an aperture of 40 mm diameter and a focal
length of 100 mm. Both simple and compound scans
had been made with this instrument. The ultra-
sonic breast scan pattern depended on the age of
the patient. Cysts could easily be demonstrated,
and so could some parts of the ductal system.
Robinson mentioned eight features, such as the
degree of shadowing, for which scans of solid
lesions required to be examined in reaching a
diagnosis. It was easier to see skin dimpling
associated with malignancy in immersed breasts,
because of the extra support provided by the water.
The possible role of ultrasonic computed
tomography in obtaining data on tissue properties
was emphasized by J. F. Greenleaf. In any particu-
lar breast, it should be possible to measure the
attenuation and velocity in the constituent image
pixels, and these data could be used to correct
conventional two-dimensional pulse-echo scans.
G. H. Glover had previously talked about his work
on ultrasonic computed tomography in a paper
presented during Session 6. As time was short.
Glover did not add to the information which he had
already given concerning the increase in velocity
associated with malignancy in breast tissue. For
the same reason. Wells did not elaborate on the
potential of neovascularization blood flow Doppler
detection, as he had mentioned this in the paper
which he had already presented in Session 4.
Melvin Linzer emphasized that computed
reflection and transmission tomography coupled
with ray tracing and frequency-dependent time-gain-
compensation techniques would be needed to make
ultrasonic breast screening feasible. Because
of the complexity of these calculations and the
vast quantities of data which would be obtained
in scanning only one patient, ultrafast processers
with large memory are required. He felt that the
potential advantages of ultrasonic investigation,
and especially the possibility of safe serial
surveys, justified the expenditure of substantial
resources on research.
12
Members of the Panel and the audience then
began an open discussion. M. L. Skolnik asked
whether it was possible to demonstrate microcalci-
fication on ultrasonic scans, and Dale replied
that this was so if the calcification was suffi-
ciently extensive. Fry pointed out that the echo-
graphy was very helpful in monitoring young women
with "lumpy" breasts, because a suggestion of the
development of malignancy could be discovered
early on; Robinson agreed with this. Pomerance
asked Fry whether she felt that mammography had
anything to contribute where a lesion had been
detected by ultrasound, and although she thought
that this might occasionally be the case, Weiss
did not. C. R. Hill returned to the value of
echography in managing patients with benign lesions,
but he felt that the method was not helpful if the
lesions were very small. Greenleaf stressed the
importance of data obtained from time-of-f 1 ight
computed tomography. Fry considered that past
progress more than justified continued efforts to
develop an ultrasonic breast screening method; but
there was general agreement with Baum when he said
that the presently available commercial equipment
ought not to be used in a trial of ultrasonic
breast cancer screening as this would inevitably
bring the method into disrepute. F. E. Barber
enquired about the potential of low-dose mammog-
raphy, and Doppman replied that the accuracies of
such techniques had not been assessed; he thought
that it would be better to wait for a viable ultra-
sonic method, than to extend the use of mammography.
According to Pomerance, even if the x-ray exposure
were to be reduced to 2 mGy, mammography would be
unacceptable for mass screening. A member of the
audience asked whether it was the tumor which was
demonstrated by echography, or the reaction of the
surrounding tissue. Robinson replied that both
features might be seen; for example, besides the
echoes from the tumor, there might be skin thick-
ening and shadowing. In this connection. Fry
mentioned the change in the collagen content of
the entire breast, which might be due to a malig-
nant tumor. Wells closed the discussion with a
short summary.
He said that it seemed to have been estab-
lished that mass screening of selected groups of
the female population for breast cancer ~ those
specially at risk, and those more than 50 years
old -- would be worthwhile, if a suitable technique
could be developed. Such a technique might well
be based on ultrasound but it would not necessarily
be a simple pulse-echo method. It would need to
be accurate and economic. The present lines of
research were promising; more resources should be
devoted to solving the fundamental and technologi-
cal problems, so that earlier clinical application
would be possible.
P. N. T. Wells
July, 1977
13
INTRODUCTORY ADDRESS
THE RELATIONSHIP OF TISSUE SIGNATURE
RESEARCH TO IMPROVED HEALTH CARE
Alfred J. Eggers , Jr.
Assistant Director for Research Applications
National Science Foundation
Washington, DC 20550
It is a pleasure to be here today before such
a distinguished audience. It has been only two
years since the first conference devoted to ultra-
sonic tissue characterization. Much has been
accomplished in those two years. Even more will
be done in the future. You should be proud to be
part of such a vigorous research effort, and I
know you share my pleasure at the prospects for
the improved medical devices that should derive
from the efforts of all of you.
It is because of these prospects that the
applications side of the National Science Founda-
tion is so interested in your efforts. In addi-
tion, there is the revolution in microelectronics
which has provided an order of magnitude reduction
in cost and improved performance every five years
for well over 25 years. Our technologists confi-
dently expect another decade of the same. What
does this mean? Suppose a highly sophisticated
piece of equipment using the current tissue signa-
ture technology would cost $1,000,000 on the
market today. I'm told that this is a reasonable
ball park estimate. Five years from now the cost
could plummet to $1 00,000--sti 1 1 a bit rich for
the average physician. But in another five, we
find that $100,000 may drop to $1 0,000--much
closer to what can be afforded.
Reflect for a moment on what the implications
of these changes could be. We may be approaching
the end of the "visible" computer because micro-
electronics can make a computer so small and
inexpensive that it will disappear behind the
imaging screen. One question is how will we
exploit this capability? This is why efforts such
as this conference are so important. It is one
crucial step in preparing wisely for the future.
Of course, we must be aware that extrapola-
tion is sometimes a hazardous experience. There
is no assurance that wanting or expecting a major
cost--performance breakthrough will make that
change happen. Still, we would all be derelict in
discharging our responsibilities if we were not
ready. After all, the sums being devoted to
research are not trivial, but they are really a
low cost insurance payment if the predicted gains
from electronics are realized. The real waste
would be to have all this great technology avail-
able and not be ready to apply it for improved
health care. You might be interested in the back-
ground for the Research Applied to National Needs
(RANN) participation with the National Bureau of
Standards and the National Institutes of Health in
support of this conference. Early in the history
of the applications directorate, we set up a
series of experiments to investigate ways of
accelerating the use of technology. One is of
particular interest to you. The first step was to
set up a study group in 1973 which evaluated the
state of development of medical ultrasonic diag-
nostic instruments. This survey team found that
exi sti ng technology could be marshalled into
improved systems. Our group of RSD incentives
experiments, therefore, included an Experiment No.
5 to encourage the industrial community to create
these new systems. As in most experiments, we
have learned and did not do everything to per-
fection. Overall, however, it was a useful
investigation. Today there is new equipment on
the market and clinical ultrasound is rapidly
becoming an important part of diagnostic services.
That same report of 1973 also had another
major finding. To make a major step forward
beyond the experiment five objectives would take
far more research across a broad front of acti-
vity. Later, we had the Alliance for Engineering
in Medicine and Biology suggest in their report
entitled, "A Five Year Research and Development
Agendum for Ultrasonic Imaging Diagnostic Instru-
mentation," that a complete system look across 23
research categories was needed. In broad terms,
the report pointed out the need to balance the
total program and consider it as a system of
interrelated parts. Funds for research on bio-
hazard were described as needed. Still, it is
silly to not avail oneself of the opportunity to
acquire information at lower powers which are
intuitively less risky. Both types of efforts
were recommended. Included in this list was
research on tissue characteri zati on--the subject
of this conference.
A direct outgrowth of the last tissue signa-
ture conference was a recognition that ultrasound
tissue signature research was not yet a coherent
body of knowledge. The RANN program provided
grant funds to establish and disseminate a pre-
liminary data base for ultrasound tissue charac-
terization research. I understand that a news-
letter is being published and that several of you
in this conference will meet on June 16 to con-
tinue the task of systematically collating re-
search results. I also understand that the first
meeting will take the place of an advisory com-
mittee under the leadership of Dr. McKinney, past
15
president of the AIUM and professor of neurology
at the Bowman Gray Medical School . Both of these
groups are part of a major effort whose objective
is clear. We want improved patient care. Moving
from the present to the future is difficult, and
we must be sure that all opinions are brought to
focus on this objective.
Now, let me point out that we have three
basic objectives in the RANN program. These are
to:
Increase the effective use of science and
technology in dealing with national
probl ems ;
Shorten the lead time between basic
scientific discoveries and relevant
practical applications; and
provide early warning of potential national
problems and initiative assessments and
research useful in avoiding or solving
such problems.
It is clear that your research efforts sat-
isfy these criteria. You will note that we em-
phasize national needs. I believe that each
country shares our concerns about improved health
care. It is a pleasure to join with all of you
from many nations to present, hear, and discuss
common research interests in this area. I wish
you all the best in your efforts today, tomorrow,
and in the future to use science and technology
more effectively to make this world a healthier
place in which to live and thrive.
16
CHAPTER 2
ATTENUATION AND VELOCITY I: MECHANISMS
17
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed., National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
ELEMENTS OF TISSUE CHARACTERIZATION
Part I. Ultrasonic Propagation Properties
R. L. Johnston, S. A. Goss, V. Maynard, J. K. Brady,
L. A. Frizzell, W. D. O'Brien, Jr., and F. Dunn
Bioacoustics Research Laboratory
University of Illinois
Urbana, Illinois 61801, U.S.A.
Tissues can be characterized ultrasonical ly by their attenuation, absorption, and
velocity, all of which correlate well with the presence of the major tissue com-
ponents of water, and protein, particularly, collagen. This correlation is examined
in solutions of biologically important molecules and in a number of tissues and
organs. It is shown that tissues can be grouped according to similar ultrasonic
propagation properties, physiological functions, and concentration of elementary
constituents. The role of collagen in determining ultrasonic properties of normal
and pathological tissues is discussed.
Key words: Absorption; amino acids; attenuation; frequency; mammalian tissues;
polypeptides; proteins; tissue characterization; ultrasonics; velocity.
1. Introduction
The ultrasonic propagation properties of bio-
logical materials include the behavior of those
measurable acoustic parameters, as functions of
the state and acoustic variables, which charac-
terize the fate of acoustic signals propagating
within the biological environment. The ultra-
sonic attenuation includes not only the absorp-
tion of the ultrasonic signal, which is degraded
to heat, but also losses due to other mechanisms
by which energy is extracted from the propagating
wave or is redirected by virtue of the inhomoge-
neous nature of the media. The ultrasonic velo-
city and the characteristic acoustic impedance,
which can be determined with the addition of
density information, embody within them both the
inertial and restoring parameters of the particu-
lar materials. Thus, knowledge of the ultrasonic
velocity and loss terms may provide a basis for
developing tissue signatures for various bio-
logical materials.
The paper will deal with the ultrasonic propa-
gation properties of tissues beginning with the
elemental constituents. As water, soft tissues,
and organs have very much the same densities and
compressibilities, it is instructive to begin a
review with properties of aqueous media.
2. Biological Molecules in Solution
A. Water
The measured ultrasonic absorption in water is
proportional to the square of the frequency, over
the range 10"^ to 10"^ MHz, with the frequency-free
absorption coefficient, a/f^, having a constant
value of 15.7 x iQ-i^ s2/cm at 37 °C. The magni-
tude of this absorption is greater than one would
expect from consideration of the so-called clas-
sical absorption due to viscosity and thermal
conductivity. This absorption in excess of the
classical value has been attributed by Hall [1]^
to a structural relaxation mechanism involving a
transition between two possible quasi-crystalline
states for water. More recent experimental re-
sults [2] are consistent with the hypothesis that
water undergoes a structural relaxation charac-
terized by a time constant of lO'^^ seconds, and
supports the idea that water is a mixture of two
or more states and that the relaxation processes
consist of the independent jumping of molecules
from one state to another.
The velocity of sound propagation in pure water
exhibits a maximum at 75 °C due to the existence
of a minimum in the product of density and adia-
batic compressibility at that temperature [3].
Similar behavior is exhibited by dilute aqueous
solutions, although the temperature of maximum
velocity may be decreased since the solutes modify
the structural arrangements of water. The veloc-
ity in water has been measured most precisely by
McSkimin [4], Greenspan and Tschiegg [5], and
DelGrosso and Mader [6].
B. Amino Acids
Since it has been observed that proteins play a
dominant role in the absorption properties of tis-
' Figures in brackets indicate literature
references at the end of this paper.
19
sues, aqueous solutions of amino acids and poly-
peptides require attention, for completeness.
When dissolved in water without additional ionic
constituents to influence the state of charge of
the amino acid, the solution exhibits a magnitude
of the frequency-free absorption parameter, a/f^,
which varies little with frequency in the mega-
hertz range, and differs only slightly from that
of the solvent, water [7-12]. Amino acids in
aqueous solution at neutral pH may be considered
according to their action of structure making or
structure breaking in the solvent. Herein, the
amino acids are present as doubly charged molecules
(zwitterions ) and are susceptible to dissociation
and recombination reactions upon a change of their
environment. Hydrogen-bonding sites are located
on both the amino and carboxyl groups, while the
side chain may be acidic, nonpolar, or basic.
Thus, the potential for breaking or making struc-
ture, in the vicinity of the solute molecules are
considerable. Hammes and Pace [7] suggested that
the predominate ultrasonic absorption mechanism in
aqueous solutions of glycine, diglycine, and tri-
glycine is that which involves solute-solvent
(water) interaction. When the pH of an amino acid
aqueous solution is within the range of 2 to 4 or
11 to 13, the relaxational behavior can be describ-
ed by a single relaxation frequency. A number of
amino acids have been investigated as a function
of pH, viz. , glycine [9,13-16], serine and threo-
nine [8], glutamic acid, aspartic acid and alanine
[14], and arginine and lysine [10,15]. Absorption
maxima have been observed within the pH ranges
2 to 4, and 11 to 13, with such peaks being de-
scribed quantitatively by assuming that the proton-
transfer reaction dominates the absorption.
C. Polypeptides
When the amino acids are formed into polypep-
tide chains, the absorption increases dramatically,
and the mechanisms believed to be responsible for
the absorption in tissues may begin to appear. In
general, aqueous solutions of polypeptides at
neutral pH exhibit an ultrasonic absorption behav-
ior greatly increased over those of amino acids in
solution or of water, and with a somewhat lesser
than the square-of-f requency dependence. The
absorption in aqueous polypeptide solutions may
involve any or all of the following four possible
mechanisms, vi z. , proton transfer, helix-coil
transition, solvation, and relaxation of the shear
viscosity. Major interest has tended to be focused
on the proton-transfer reactions [17-20] and
helix-coil transitions [21-24].
D. Proteins
Continuing with increasing complexity of the
biological media, the polypeptides may be con-
sidered to be arranged in a particular, -ubiquitous
way to form the globular proteinaceous state.
Hemoglobin and serum albumin solutions have re-
ceived considerable attention, partly because of
the availabilities of the materials. Special con-
sideration is given to other macromolecul ar species
such as nucleic acids and polysaccharides.
When biological macromolecules , such as the
globular protein hemoglobin are in aqueous solution,
a certain amount of the solvent becomes an inherent
part of the molecule since the polymer possesses
ionic and polar groups which associate with water
molecules. In addition, proteins contain a number
of nonpolar side chains such that, within the
vicinity of the macromolecules, some water struc-
turing occurs. Thus, it is possible that the
structure of liquid water, the hydration layer,
increases in the neighborhood of the biological
macromolecule. It is considered, therefore, that
as an acoustic wave propagates through an aqueous
solution of biopolymers, it perturbs the hydration
layer manifesting absorption of energy by a struc-
tural relaxation process. The role of molecular
conformation of the biopolymer has also been con-
sidered as the origin of the observed ultrasonic
absorption. Figure 1 shows the excess frequency-
free absorption per unit concentration for several
biomacromolecules and supports the view that struc-
turing contributes to the ultrasonic absorption
spectra [25]. Both dextran [26], a carbohydrate,
and polyethylene glycol [27], a synthetic polymer,
assume random coil configurations in aqueous solu-
tion and exhibit absorption magnitudes similar to
that of gelatin. Hemoglobin has a quaternary
structure, bovine serum albumin and ovalbumin have
tertiary structure, polyglutamic acid [28], a syn-
thetic polyamino acid, has secondary structure,
and the double helix, DNA, has a rigid rod con-
formation [29]. However, hemoglobin in 5 molar
aqueous guanidine hydrochloride solution exists
as a random coil [30] and yet exhibits ultrasonic
absorption spectra similar to that of hemoglobin
in aqueous solutions [31]. The importance of
molecular spatial arrangement is thus an unsettled
question, though the polypeptide structure appears
to be of considerable significance.
1 1 1
1 1 ' 1 1 1 1 M 1.
^DNA-SS
10 °c :
^BSA-
OX,IO°C
/ \ x>0^\
/ ^ >
/ voO^
/ vO-
- Hb-0X,IO°C
^ .OM
-ox,io°c
f DNA-CT,I0°C^"^ ^
^PGA pH 5.8 ,25°C -_
■^^ DEXTRAN, 20°C -
- PEG, 20.7 "C^
/Hb,25°C
PEG, 4.2°C
,1
1 1^ t 1 1 1 1 1 1
ll I I l^l I 1 1 1 1 1 I
10 100 1000
frequency ( MHz )
Fig. 1. Excess frequency-free ultrasonic absorp-
tion per unit concentration [25].
Within figure 1, bovine serum albumin and hemo-
globin have a molecular weight of 68,000 while
that of ovalbumin is 46,000. e-lactoglobin [32]
(not on graph) with a molecular weight of 35,000,
would fall in the range determined by hemoglobin,
bovine serum albumin, and ovalbumin, while lyso-
zyme [32] (not on graph) with a molecular weight
of 14,600, would appear between hemoglobin at
25 °C and dextran. Two random coil polymers,
dextran and polyethylene glycol, have been studied
as a function of molecular weight. For dextran
solutions, the frequency-free absorption per unit
concentration increases with increasing molecular
weight to a molecular weight of about 10,000,
which corresponds to approximately 100 monomer
units, beyond which it is independent of molecular
20
weight [33]. Aqueous polyethylene glycol solu-
tions show similar absorption behavior in that be-
yond a molecular weight of about 4500, which also
corresponds to a chain length of about 100 monomer
units, the absorption is independent of molecular
weight [27]. Thus, ultrasonic absorption depends
to some degree upon molecular weight. Possibly
beyond about 100 monomer units, the macromolecule
assumes random coil characteristics.
It is uncertain whether the absorption mechan-
ism(s) responsible for energy loss in biological
polymer solutions (usually of concentrations less
than 10 percent by weight) are the same as those
responsible for the absorption properties of tis-
sue [34]. Kremkau and Carstensen [11] have sug-
gested that more highly concentrated macromolecular
solutions, promoting solute-solute ( intermolecul ar )
interactions, may better approximate the tissue
environment, and from the point of view of biologi-
cal effects, may be the most important level at
which the ultrasound interacts. In the case of
dilute solutions, where the absorption is con-
sidered to vary linearly with polymer concentra-
tion, the mechanism of absorption is attributed to
processes involving the interaction between the
solvent (usually water) and the solute. In other
than dilute solutions, the absorption is found to
increase nonlinearly, and it has been suggested
that this phenomenon is due to intermol ecul ar in-
teraction processes [34-36]. It has recently been
shown [37], from absorption measurements of bovine
serum albumin, to concentrations of 40 g/100 ml,
over the frequency range 3.4 to 15 MHz, and at
20 °C, that the absorption dependence upon con-
centration does not possess a linear region.
Here, the excess ultrasonic absorption can be de-
scribed adequately by a concentration dependence
to the 1.2 power. It thus appears essential to
consider intermolecular interaction over the en-
tire range.
3. Tissues
The nearly linear frequency dependence of at-
tenuation in liver, kidney, brain, muscle, fat,
and other parenchyma, are considered in the ap-
proximate frequency range of 100 kHz to 100 MHz,
and the recent findings with regard to the tem-
perature dependence are discussed. Several tis-
sues and organs, such as bone, lung, refractive
media of the eye, and collagenous tissues, are
singled out for special detailed consideration.
It is shown that classification of tissues accord-
ing to certain ultrasonic propagation properties
can be carried out with regard to water and col-
lagen content, and with regard to certain teleo-
logical considerations. Pulmonary tissue is an
exception here.
A. Frequency Dependence
The dependence of the ultrasonic absorption co-
efficient upon the acoustic frequency has been
studied by numerous investigators. Goldman and
Hueter [38] prepared an early compilation of ultra-
sonic velocity and absorption data in mammalian
soft tissues. Therein it is seen that the veloci-
ty, excluding lung, is very nearly that of dilute
salt solutions and varies only slightly among
those tissues. Fatty tissues are exceptions in
having a velocity about 10 percent less than the
others. Figure 2, taken from their paper, is a
graphical representation of the acoustic amplitude
absorption coefficient per wavelength for several
mammalian tissues in the frequency range of ap-
proximately 200 kHz to 10 MHz. As they attempted
to include all measurements available at that time,
by numerous investigators employing different ex-
perimental techniques, the scatter of the data
exhibited by the bands, or broad shaded regions is
thus not wholly surprising since many neglect to
give either complete specifications of their ex-
perimental procedure or a description of the state
of the specimen used. For example, it is not pos-
sible, in many cases, to determine from the litera-
ture the temperatures employed by the investigators
reporting the data. It is, however, possible to
discern several relatively simple relationships.
For example, the absorption per cycle, a/f, is
generally constant over the frequency range con-
sidered. For fat, a/f increases slightly in the
frequency range from 1 to 10 MHz. The experiment-
al results for striated muscle and liver appear to
exhibit a minimum in the neighborhood of 2 MHz.
Kessler [39] has shown that for kidney the linear
dependence of the attenuation on frequency exists
to about 100 MHz, after which a square law (or
greater) dependence exists. Fry [40] has con-
sidered a viscous mechanism for the absorption of
ultrasound in tissue, in which it is shown that
the viscous forces acting between a suitably chosen
distribution of suspended particles (or structural
elements) and a suspending liquid can account for
the experimentally observed linear relationship be-
21
tween acoustic absorption coefficient and ultra-
sonic frequency. The frequency band over which
linearity obtains (in the model) is determined by
the limits of the distribution Of values for the
parameters chosen to describe the structural ele-
ments. Below the linear range the theory predicts
a quadratic dependence, in agreement with ex-
periment.
Figure 3 is another way of presenting the data
which may be suggestive for determination of
mechanisms of absorption [41]. Here, the loga-
rithm of the absorption coefficient is plotted as
a function of the logarithm of the ultrasonic fre-
quency, and the slopes of the resulting curves are
examined (the slope is the exponent of frequency
upon which the magnitude of the absorption coeffi-
cient depends). Figure 3 shows data for several
materials of increasing biological complexity,
exhibiting correspondingly increasing complexity
in absorptive behavior. The urea solution exhibits
a slope of 2, indicative of classical viscous
absorption for which a/f^ is a constant. Homogen-
ized milk, a suspension of fat particles and
hydrated casein complexes, exhibits a slope of
nearly unity from approximately 1 to 40 MHz. Be-
havior of this type cannot be explained in terms
of simple viscosity or scattering theories. The
curves of the absorption coefficients of egg al-
bumin, brain tissue, liver, and striated muscle
(not shown in fig. 3) exhibit slopes between 1 and
2 in the neighborhood of 1 MHz and approach a slope
of 2 at higher frequencies. Hueter [41] has sug-
gested that this type of frequency dependence can
be described for specific muscle preparations by
a double relaxation process in which the bulk
(volume) viscosity of the tissue possesses a relax-
ation frequency near 40 kHz, and the shear viscosity
possesses a relaxation near 400 kHz. Although it is
conceded that this is an oversimplification of a
complicated process, it will be shown below that the
temperature dependence of the acoustic absorption
coefficient lends support to this view.
m 1.0
-o 8
■S 6
I I
§ 3
c
t 2
0.1
0.02
1 1 1 1 1 1 1 1
whole liver, -^
1 hour \
fresh brain —
1 1 1 1 1 y 1
// / /
/ /
/ // /
1 //^XT\\\Y. /
/ '// '
/ /
/ / -
/ ^ ^ _
liver homog
1 hour /
/ // — y- —
/A /
// /
/''^ ^ egg white /
lOm.urea^ /
1 / 1 1 1 1 1 1
1 1 1
0.1
1.0
10
*50
frequency in (MHz)
Fig. 3.
Acoustic amplitude absorption coefficient
versus frequency for materials of differ-
ent biological complexity [41].
B. Temperature Dependence
Details of the absorption coefficient as a
function of temperature and frequency have re-
cently become available. Figure 4 shows observa-
tions on mammalian central nervous system, the
only tissue for which such data are available.
0.8
O.G
e
o
0.2
0.26 MHz
0.5 MHz
0.7 MHz
"^'^^ 4.2 MHz
Fig. 4. Frequency and temperature dependence of
ultrasonic absorption in mammalian central
nervous tissue [43].
The curves for 0.26, 0.5, 0.7, and 1 MHz rep-
resent in vivo measurements in the spinal cords
of neonatal mice (essential poi ki 1 otherms )
[42-44] and those for 4.2 MHz are in vitro
measurements in brains of adult cats (homeo-
therms) [45]. The relatively complex behavior of
the frequency-free absorption coefficient with
frequency and temperature suggests a family of
curves whose maxima decrease in magnitude, and oc-
cur at ever higher temperatures, as frequency in-
creases, supporting the suggestion mentioned
above. It is not known whether other soft tis-
sues exhibit similar behavior but Kishimoto [46]
has observed a positive temperature coefficient
for the absorption coefficient of bone in the
frequency range 1.4 to 4.5 MHz. These data il-
lustrate the necessity for complete specification
of the state of specimens when reporting experi-
mental results.
C. Absorption and Velocity in Bone
Bone is a tissue possessing acoustic propaga-
tion properties greatly different from those of
the soft tissues discussed previously. An early
study of specially prepared skull bone, in the
frequency range 0.6 to 3.5 MHz (25 to 35 °C),
yielded a quadratic dependence of the absorption
coefficient upon frequency with a transition to a
linear dependence beyond about 2 MHz [47]. An
average value found for the acoustic amplitude
22
absorption coefficient per unit path length in
skull bone, in the neighborhood of 1 MHz, was of
the order of 1 cm"^, approximately an order of
magnitude greater than that of soft tissues of the
same temperature. However, recent observations
have called these values into question, and Adler
and Cook [48] have obtained absorption measure-
ments of 1.5 cm"^ and 2.2 cm"^ in freshly frozen
dog tibia at room temperature at, respectively,
3 and 5 MHz. Reports of measurements of the longi-
tudinal speed of sound in bone are largely in
agreement that it is approximately twice that of
soft tissues [46,48-50]. Anisotropy of elastic
properties and variations in density of bone
present special problems for measurement and for
interpretation of results [48-50].
D. Refractive Media of the Eye
Begui [51] has studied the acoustic properties
of the refractive media of the eye in vitro. He
determined the ultrasonic absorption coefficient
of the aqueous and vitreous humors at 30 MHz and
that of the lens at 3 MHz. The specimens were ob-
tained from excised fresh calf eyes. At 30 MHz
and 27.5 °C, the aqueous and vitreous humors both
exhibit an acoustic amplitude absorption coeffi-
cient of 0.35 cm"^. Since this is approximately
50 percent greater than the absorption coefficient
of dilute salt solutions, it suggests that the ab-
sorption coefficients of the humoral media of the
calf eye possess a viscous-type dependence upon
frequency; that is, the absorption coefficient
probably increases as the square of the frequency.
The lens of the calf eye exhibits a value of 0.7
cm"-' for the acoustic amplitude absorption coef-
ficient at 3 MHz and 28 °C. Since the lens con-
tains a relatively high concentration of protein,
it is reasonable to assume, in the absence of
further information, that the frequency dependence
of the absorption coefficient of the lens re-
sembles that of other soft tissue for which the
absorption appears to be dominated by the protein
content, i.e., it is probable that the absorption
coefficient per unit path length of the lens
varies approximately with the first power of the
frequency. Some investigators currently using
ultrasonic methods for diagnosing disorders of
the human eye feel that the lens absorption value
given by Begui is larger than that for the human
lens in vivo. The possible discrepancy may result
because of species differences. Indeed, Begui
observed that the viscosity of the intraocular
fluid of calf eyes is greater than the values
normally stated for the fluid media of human eyes.
Further, the specimens used by Begui were first
stored (at temperatures in the neighborhood of
0 to 5 °C) and were used for measurement purposes
within a time interval of 10 days. Begui obtained
for the speed of sound in refractive media of the
eye 1497 m/s for the aqueous humor, 1516 for the
vitreous humor, and 1616 for the lens.
E. Pulmonary Tissue
Two earlier studies [52,53] showed that ultra-
sonic attenuation in freshly excised dog lung was
unusually high, that the speed of sound was con-
siderably less than that of water, and that both
of these quantities had a strong dependence on
pulmonary inflation and acoustic frequency. It
was also shown that a pathological condition in-
volving an accumulation of liquid-like matter
within the pulmonary architecture, had the effect
of appreciably reducing both the attenuation and
the velocity. Two recent studies have provided
details of the frequency and inflation dependencies.
Dunn [54] has shown that for excised dog lung of
inflation to a fraction of residual air, such that
the specimen density is 0.4 g/cm^, the attenuation
increases exponentially from 4 cm"^ at 1 MHz to
12 cm~i at 5 MHz. In this same frequency range,
the speed of sound increased linearly from 0.66 x
10^ cm/s to 1.2 X 10^ cm/s. These findings are in
general agreement with those of Bauld and Schwan
[55] who also showed that, for fixed inflated
specimens, the energy reflected at the lung-liquid
interface ever increases the gaseous inflation
allowing for lesser amounts of energy to enter
the lung.
4. Role of Collagen
Collagen is the most abundant single protein in
the human body and the most common protein in the
animal kingdom. It is closely associated in con-
nective tissue of vertebrates and comprises be-
tween one-quarter and one-third of the total
protein in the human body, being about six per-
cent of the total body weight [56]. However, more
than the prevalence of collage in the body, there
is some evidence to suggest that its contribution
to the elastic properties of most soft tissues,
together with other structural proteins, deter-
mines acoustic contrast during echographic visuali-
zation [57,58]. This hypothesis is based on the
fact that the static or low- frequency elastic
modulus of collagenous fibers is at least 1000
times greater than those of soft tissues. Since
the ultrasonic velocity is proportional to the
square root of the elastic modulus, collagenous
tissues are thought to introduce a greater im-
pedance mismatch than would be the case for a soft
tissue interface, thereby increasing the acoustic
reflectivity. The increased deposition of collagen
and the concomitant increase in attenuation seen
in many pathological conditions is a basis for
ultrasonic differential diagnosis.
Table 1 contains ultrasonic attenutati on , ve-
locity, water content, total protein content, and
collagen content for various tissues [59]. It is
apparent that the greater the collagen content,
the greater the attenuation. Dependence of at-
tenuation and velocity upon water content are al-
so apparent. These data allow empirical relations
to be formed for more quantitative assessment of
the role of collagen content of tissues upon their
ultrasonic propagation properties. For example,
figure 5 represents a summary of table I of the
ultrasonic attenuation at 1 MHz as a function of
the wet weight percentage of collagen for ten tis-
sues. Using linear regression by the method of
least squares, a reasonable fit to the data is
provided by the relation
A = 0.17 C°-3, (1)
where A is the ultrasonic attenuation in cm"^ and
C is the wet weight percentage of collagen. The
best fit parameter, the coefficient of determina-
tion, r^, yields a value of 0.71 (unity represents
a perfect fit). Equation 1 is represented on
figure 5 by the solid straight line. Logarithmic,
exponential, and linear functions were also analyz-
23
Table 1. Ultrasonic attenuation and velocity for tissues of various
water, protein and collagen content.
Tissue Attenuation at Velocity Water Protein Collagen
1 MHz (cm-i) (m/s) (%) (%) (%)
Water (20 °C)
0.0003
1483
100
Amniotic fluid
0.0008
1510
97
0.27
Agueous humor
0
10 - 0.017
QQ
u . UU J - 1
Vitreous humor
0
10 - 0.017
1 D 1 0
QQ
yy -
QQ
yy
y
U . - U . £13
0.014 -
0.06;
r c r
Lor
0.0012
1 A QQ
1 0 1 D
QQ
yy
(J, UJ
PI a sma
0.01
1 0/ 1
y u -
QR
y D
7
Tes ti s
0.019
(absorption)
Q/l
trace
Blood
0.02
1571
74 -
83
Milk
0.04
1485
87
3 - 4
Fat
0
04 - 0.09
1410 -
1479
10 -
19
5 - 7
yes
Spl een
0.06
1520 -
1591
76 -
80
17 - 18
0.5
-1.2
Liver
0
07 0.13
1 550 -
1607
68 -
78
20 - 21
0.1
-1.3
Kidney
0
09 - 0.13
1558 -
1568
76-83
15 - 17
0.5
- 1.5
Brain
0
09 - 0.13
1510 -
1565
75 -
79
10 10
0.04
- 0.3
Spinal cord
0
09 - 0.12
64 -
80
Striated muscle
against grain
with grain
0
0
18 - 0.25
08 - 0.12
0.16
1 DDO -
1592 -
1 576 -
1 cnQ
1 oUi
1603
1587
66 -
80
on 91
0.7
- 1.2
Heart
0
25 - 0.38
1572
77 -
78
17
0.4
- 1.6
Tongue
against grain
with grain
0.58
0.28
1 ETC
1 0/0
1585
62 -
68
14 - 17
Lens
0
10 - 0.20
1616
63 -
69
30 - 36
Articular capsule
0.38
Integument
0.40
1498
60 -
72
7
- 30
Cartilage
0.58
1665
23 -
34
70
49 - 63
10
- 20
Tendon
against grain
with grain
0.54
0.58
1750
63
35
32
0.01
tissue collagen (percentage)
Fig. 5. Attenuation at 1 MHz as a function of the percentage of tissue
collagen for 10 tissue types [59].
24
ed but yielded worse fits than equation 1. Simi-
larly, to a first approximation, the ultrasonic
velocity was examined as a function of collagen
content for 8 tissues, excluding integument, and
yielded the expression
C = -1700 + 230 In v, (2)
where v is the ultrasonic velocity in meters/second
and r^ = 0.91. Again, first approximations for wet
weight percentage of total protein, P, yielded
A = 0.004 pi-26, r2 = 0.77, (3)
for 16 biological materials excluding integument.
Thus to a first approximation there appear to be
mathematical relationships which can be developed
to relate the amount of tissue constituents to the
ultrasonic propagation properties. There are some
tissues, such as fat and integument, which may
have to be treated separately but, otherwise, this
approach suggests that such relationships may aid
in developing ultrasonic tissue signatures which
can be incorporated into clinical instrumentation.
5. Concluding Remarks
What emerges from all this is summarized in
table 2, following Dussik and Dunn [50,61], which
is an attempt to characterize tissues according
to their ultrasonic propagation properties and
biological function. It is seen that tissues can
be grouped according to apparent teleology fashion
with relatively narrow ranges of attenuation
values within each group. The attenuation approxi-
mately doubles from group to group in the direc-
tion of increasing attenuation, and the speed of
sound increases in the same direction. Further,
proceeding from group to group in the same direc-
tion, tissues of ever-decreasing water content and
ever-increasing structural protein content become
included. Thus, it is seen that ultrasonic at-
tenuation and velocity may be invoked to charac-
terize tissues according to functional, structural,
and teleological criteria. Possibly detailed mea-
surements will allow assignment of resolvably
unique values to each tissue structure, including
usefully differentiable values for pathological
states. Should this be the case, ultrasonic
attenuation and impedance values, as functions
of state and acoustic parameters, media, etc . ,
should specify uniquely tissues for diagnostic
purposes .
Acknowledgment
The authors acknowledge gratefully the partial
support for portions of the activities described
herein by grants from the National Institutes of
Health.
Table 2. Average attenuation of tissues by categories.
Tissue
attenuation
categories
Attenuation
at 1 MHz
(cm-i)
Ti ssue
Assumed
teleology
General
trends
1. Very low
2. Low
3. Medium
4. High
5. Very high
0.03
0.01
0.06-0.07
0.08-0.11
0.11
0.08-0.16
0.23
0.3
0.4
0.5
0.6
1 or more
> 4
serum
blood
adipose tissue
nervous tissue
1 i ver
muscle
heart
kidney
integument
tendon
carti lage
bone (mineralized)
pulmonary tissue
ion , metabol ic , etc . ,
transport convection
energy and (water)
storage
physiological
function
parenchymal
tissue
structural
integration
stromal tissues
skeletal framework
gaseous exchange
Increas- Increas-
ing ing
struc- speed
tural of
protein sound
content
Increas-
ing
H2O
content
References
[1] Hall, L., The origin of ultrasonic absorp-
tion in water, Phys. Rev. 73, 775 (1948).
[2] Davis, C. and Jarzynsky, J., Liquid Water-
Acoustic Properties: Absorption and Re-
laxation, Chapter 12 in The Physics and
Physical Chemistry of Water, Vol . I of Water:
A Comprehensive Treatise, F. Franks, ed.,
(Plenum Press, New York, 1972).
[3] Willard, G. W. , Temperature coefficient of
ultrasonic velocity in solutions, J. Acoust.
Soc. Am. 19, 235 (1947).
[4] McSkimin, H. J., Velocity of sound in distil-
led water for the temperature range 20 °C -
75 °C, J. Acoust. Soc. Am. 37, 325 (1965).
[5] Greenspan, M. and Tschiegg, C. E., Speed of
sound in water by a direct method, J. Res.
Natl. Bur. Std. 59 249-254 (1957).
[6] DelGrosso, V. A. and Mader, C. W. , Speed of
sound in pure water, J. Acoust. Soc. Am. 52 ,
25
1442-1446 (1972).
[7] Hammes, G. G. and Pace, C. N., Ultrasonic
absorption measurements in aqueous solutions
of glycine, diglycine, and triglycine,
J. Phys. Chem. 72, 2227 (1968).
[8] White, R. D., Slutsky, L. J., and Pattison,
S., Kinetics of the proton transfer reac-
tions of serine and theronine, J. Phys.
Chem. 25, 161 (1971).
[9] Hussey, M. and Edmonds, P. D., Ultrasonic
examination of proton-transfer reactions in
aqueous solutions of glycine, J. Acoust.
Soc. Am. 49, 1309 (1971).
[10] Hussey, M. and Edmonds, P. D., Ultrasonic
examination of proton-transfer reactions at
the a-amino and side-chain groups of arginine
and lysine in aqueous solution, J. Acoust.
Soc. Am. 49, 1907 (1971).
[11] Kremkau, F. W. and Carstensen, E. L. ,
Macromolecular Interaction in Sound Absorp-
tion, in Interaction of Ultrasound and
Biological Tissues--Workshop Proceedings,
J. M. Reid and M. R. Sikov, eds., DHEW
Publication (FDA) 73-8008 BRH/DBE 73-1,
p. 37 (U.S. Government Printing Office,
Washington, D.C. 1972).
[12] Kremkau, F. W., Macromolecular Interaction
in the Absorption of Ultrasound in Biologi-
cal Material, Ph.D. Thesis, University of
Rochester (1972).
[13] Maass, G., Ph.D. Thesis, University of
Gottingen (1962).
[14] Applegate, K. , Slutsky, L. J., and Parker,
R. C, Kinetics of proton-transfer reactions
of amino acids and simple polypeptides,
J. Am. Chem. Soc. 90, 6909 (1968).
[15] Hussey, M. , Proton Transfer at Side Chain
Groups: A Mechanism for the Absorption of
Ultrasound in Protein Solutions, Ph.D.
Thesis, University of Pennsylvania (1970).
[16] Hussey, M. , Proton Transfer Processes: One
of the Modes of Interaction of Ultrasound
with Proteins and DNA, in Interaction of
Ultrasound and Biological Tissues, J. M.
Reid and M. R. Sikov, eds., DHEW Publication
(FDA), 73-8008 BRH/DBE 73-1, p. 31 (U.S.
Government Printing Office, Washington, D.C.
1972).
[17] Zana, R. and Tondre, C, Ultrasonic studies
of proton transfers in solutions of poly-
(lysine) and poly(ornithine) . Implications
for the kinetics of the helix-cord transi-
tion of polypeptides and for the ultrasonic
absorption of proteins, J. Phys. Chem. 76
1737 (1972).
[18] Lang, J., Tondre, C, and Zana, R. , Effect
of urea and other organic substances on the
ultrasonic absorption of protein solutions,
J. Phys. Chem. 75, 374 (1971).
[19] Zana, R. and Lang, J., Effect of pH on the
ultrasonic absorption of aqueous solutions
of proteins, J. Phys. Chem. 74, 2734 (1970).
[20] Applegate, K. , Slutsky, L. J., and Parker,
R. C, Kinetics of proton-transfer reactions
of amino acids and simple polypeptides, J^.
Am. Chem. Soc. 90, 6909 (1968).
[21] Doty, P., Wada, A., Yang, J. T., and Blout,
E. R. , Polypeptides: VIII, molecular con-
figurations of poly-L-glutamic acid in
water-dioxane solution, J. Polymer Sci. 23,
851 (1957).
[22] Barksdale, A. D. and Stuehr, J. E., The
kinetics of the helix-coil transition in
aqueous poly-L-glutamic acid, J. Am. Chem.
Soc. 94, 3334 (1972).
[23] Schwarz, G., On the kinetics of helix-coil
transition of polypeptides in solution,
J. Mol. Biol. n_, 64 (1965).
[24] Wada, Y., Sasabe, H., and Tomono, M. ,
Viscoelastic relaxations in solutions of
poly-(glutamic acid) and gelatin at ultra-
sonic frequencies, Biopolymers 5, 887 (1967).
[25] Dunn, F., Acoustic Properties of Biological
Materials, in Fundamental and Applied
Aspects of Nonionizing Radiation, S. M.
Michaelson and M. W. Miller, eds., p. 21
(Plenum Publishing Corporation, New York,
1975).
[26] Hawley, A. S., Kessler, L. W., and Dunn, F. ,
Ultrasonic absorption in aqueous solutions
of high-molecular-weight polysaccharides,
J. Acoust. Soc. Am. 38, 521 (1965).
[27] Kessler, L. W., O'Brien, W. D., Jr., and
Dunn, F., Ultrasonic absorption in aqueous
solutions of polyethylene glycol, J. Phys.
Chem. 74, 4096 (1970).
[28] Burke, J. J., Hammes, G. G., and Lewis,
T. B., Ultrasonic attenuation measurements
in poly-L-glutamic acid solutions, J. Phys.
Chem. 42, 3520 (1965).
[29] O'Brien, W. D. , Christman, C. L., and Dunn,
F., Ultrasonic investigation of aqueous
solutions of deoxyribose nucleic acid,
J. Acoust. Soc. Am. 52, 1251-1255 (1972).
[30] Kawahara, K. , Kirscher, A. G., and Tanford,
C, Dissociation of human co-hemoglobin by
urea, guanidine hydrochloride and other
reagent. Biochemistry 4, 1203 (1965).
[31] O'Brien, W. D., Jr. and Dunn, F. , Ultrasonic
examination of the hemoglobin dissociation
process in aqueous solutions of guanidine
hydrochloride, J. Acoust. Soc. Am. 50, 1213-
1215 (1971).
[32] Lang, J., Tondre, G., and Zana, R., Effects
of urea and other organic substances on the
ultrasonic absorption of protein solutions,
J. Phys. Chem. 75, 374 (1971).
26
[33] Hawley, S. A. and Dunn, F., Ultrasonic ab-
sorption in aqueous solutions of dextran,
J. Chem. Phys. 50^, 3523 (1969).
[34] O'Brien, W. D. , Jr. and Dunn, F., Ultra-
sonic Absorption by Biomacromolecules ,
in Interaction of Ultrasound and Biological
Tissues--Worl<shop Proceedings, J. M. Re id
and M. R. Sikov, eds., p. 13, DHEW Publica-
tion (FDA) 73-8008 BRH/DBE 73-1 (U.S. Govern-
ment Printing Office, Washington, D.C. 1972).
[35] Kessler, L. W. and Dunn, F., Ultrasonic
investigation of the conformal changes of
bovine serum albumin in aqueous solution,
J. Phys. Chem. 73, 4256 (1969).
[36] O'Brien, W. D., Jr. and Dunn, F., Ultrasonic
absorption mechanisms in aqueous solutions
of bovine hemoglobin, J. Phys. Chem. 76 ,
528 (1972).
[37] Goss, S. A. and Dunn, F., Concentration De-
pendence of Ultrasonic Absorption in Aqueous
Solutions of Bovine Serum Albumin, in IEEE
Ultrasonics Symposium Proceedings, p. 65
(1974).
[38] Goldman, D. E. and Hueter, T. F., Tablar
data of the velocity and absorption of high-
frequency sound in mammalian tissues, J_.
Acoust. Soc. Am. 28, 35 (1956); 29, 655
(1957).
[39] Kessler, L. W., VHF ultrasonic attenuation
in mammalian tissue, J. Acoust. Soc. Am. 53
1759 (1973).
[40] Fry, W. J., Mechanism of acoustic absorption
in tissue, J. Acoust. Soc. Am. 24, 412-415
(1952).
[41] Hueter, T. F., Viscoelastic Losses in Tissues
in the Ultrasonic Range, WADC Technical Re-
port 57-706 (1958).
[42] Dunn, F., Temperature and amplitude depend-
ence of acoustic absorption in tissue, J_.
Acoust. Soc. Am. 34, 1545 (1962).
[43] Dunn, F. and Brady, J. K., Pogloshchenie
ul ' trazvyeka v biologicheski kh sredakh,
Biofizika 18, 1063 (1973). Translation,
Ultrasonic absorption in biological materials.
Biophysics 18, 1128 (1974).
[44] Dunn, F. and Brady, J. K. , Temperature and
Frequency Dependence of Ultrasonic Absorp-
tion in Tissue, in Proceedings 8th Interna-
tional Congress on Acoustics, Vol. I, p.
366c (Goldcrest Press, Trowbridge, Wilts.,
1974).
[45] Robinson, T. C. and Lele, P. P., An analysis
of lesion development in the brain and in
plastics high intensity focused ultrasound
at low megahertz frequencies, J. Acoust.
Soc. Am. 51, 133 (1972).
[46] Kishimoto, T. , Ultrasonic absorption in
bones, Acustica 8, 179 (1958).
[47] Hueter, T. F., Messung der Ultraschal labsorp-
tion in menschlichen Schadel knochen und ihre
Abhangigkeit von der Frequenz, Naturwissen-
schaften 39, 21 (1952).
[48] Adler, L. and Cook, K. V., Ultrasonic para-
meters of freshly frozen dog tibia, J.
Acoust. Soc. Am. 58, 1107-1108 (1975T.
[49] Thiesmann, H. and Pfander, F. , Uber die
Durchlassigkeit des Knochens fiir Ultraschall,
Strahlentherapie 80. 607 (1949).
[50] Lang, S. B., Ultrasonic Method for Measuring
Elastic Coefficients of Bone and Results on
Fresh and Dried Bovine Bones, IEEE Trans.
Biomed. Eng. BME-17, 101 (1970T^
[51] Begui, Z. E., Acoustic properties of the re-
fractive media of the eye, J. Acoust. Soc.
Am^ 26 , 365 (1954).
[52] Dunn, F. and Fry, W. J., Ultrasonic absorp-
tion and reflection by lung tissue, Phys.
Med. Biol. 5, 401 (1961).
[53] Reid, J. M., Ultrasonic Diagnostic Methods
in Cardiology, Ph.D. Thesis, University
of Pennsylvania, 1965.
[54] Dunn, F., Attenuation and speed of sound in
lung, J. Acoust. Soc. Am. 56, 1638 (1974).
[55] Bauld, T. J. and Schwan, H. P., Attenuation
and reflection of ultrasound in canine lung
tissue, J. Acoust. Soc. Am. 56^, 1630 (1974).
[56] White, A., Handler, P., and Smith E. L.,
Principles of Biochemistry (McGraw Hill Book
Co., New York, 1968).
[57] Fields, S. and Dunn, F., Correlation of
echographic vi sual izabi 1 i ty of tissue with
biological composition and physiological
state, J. Acoust. Soc. Am. 54, 809-812 (1973).
[58] Kessler, L. W., Fields, S. I., and Dunn, F.,
Acoustic microscopy of mammalian kidney,
J. Clinical Ultrasound 2, 317-320 (1974).
[59] O'Brien, W. D. , The Role of Collagen in
Determining Ultrasonic Propagation Proper-
ties in Tissue, in Acoustical Holography,
L. W. Kessler, ed.. Vol. 7 (Plenum Press,
New York, 1977).
[60] Dussik, K. T. , Kyriazidov, M. , Fritch, D. J.,
and Srear, R. S., Measurement of articular
tissues with ultrasound, Amer. J. Phys. Med.
37, 160-165 (1958).
[61] Dunn, F., Ultrasonic Attenuation, Absorption,
and Velocity in Tissues and Organs, in Ul tra-
sonic Tissue Characterization, M. Linzer,
ed. , National Bureau of Standards Special
Publication 453, pp. 21-28 (U.S. Government
Printing Office, Washington, D.C. 1976).
27
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed.. National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
ABSORPTION OF SOUND IN TISSUES
Edwin L. Carstensen
Department of Electrical Engineering
University of Rochester
Rochester, New York 14627, U.S.A.
In spite of extensive applications of ultrasound in diagnosis, therapy and even
surgery, there are still many problems to be solved in the basic physics of sound
propagation in tissues. Absorption of longitudinal ultrasonic waves occurs primarily
at the macromolecular level. There is evidence to indicate that this absorption can
be profoundly modified by macromolecular interaction. The specific structural or
chemical relaxation mechanisms responsible for the absorption are unknown. Microscopic
inhomogeneities may lead to certain forms of relative motion viscous losses or thermal
absorption. Macroscopic inhomogeneities in tissue affect sound propagation and can
lead to artifacts in certain methods of measurement of tissue absorption. Shear waves
are not important in the soft tissues of the body.
Key words: Absorption of ultrasound; macromolecular relaxation; relaxation phenomenon;
ultrasonic tissue absorption.
Although acoustic absorption phenomena play
an important role in every medical application
of ultrasound, we still have only a rudimentary
understanding of the responsible physical mech-
anisms forty years after the first work in this
field [1]^. This review of basic concepts at-
tempts to outline what we know about these
processes and to identify problems to be solved.
1. Absorption Data
Information on absorption of sound in tissues,
as it is available to us today, is outlined in
figure 1. (A more complete summary of absorp-
tion data is included in the paper by Johnston
et al . [2].) Data for packed red cells are in-
cluded iDecause we can think of this preparation
as a kind of simple model tissue and because it
has been studied over a wide range of frequencies.
The absorption of sound in fatty tissue is
significantly lower than the other soft tissues.
Bone is much lossier than other tissues of the
body. Water makes a negligible contribution to
the absorption of tissues at low frequencies.
Only a few measurements have been carried out
above 10 MHz. We can guess that values of (aA)
for soft tissues will follow the pattern set by
hemoglobin [3] and approach but remain signifi-
cantly above that of water at frequencies of
the order of 1000 MHz. From acoustic microscopy
it is clear that sharp differences in absorption
among components of cells exist at frequencies
above 1000 MHz [4].
The range of values which have been reported
for a given tissue type is large. This results
^Figures in brackets indicate literature
references at the end of this paper.
in part from normal biological variability. In
addition, there is a potential for error in
measurement with i nhomogeneous tissue samples.
10"
2 5 2 5 2 5 2 5
0.1 1.0 10 100 1000
FREQUENCY (MHz)
Fig. 1. Absorption of sound in biological mater-
ials. Solid lines are estimates of mean
absorption coefficients from many investi-
gators. The range of reported values is
large particularly for bone. The high
frequency extrapolation (dashed line) for
concentrated red cells is based on meas-
urements of hemoglobin solutions [3]. Ex
Exprapolation of the data for soft tissues
(dashed line) to 100 MHz is based on meas-
urements of kidney at 100 and 200 MHz
[35].
29
For example, we found that in measuring sample
attenuation with a phase sensitive receiver the
apparent absorption for muscle could be as much
as four times its true value [5]. It is inter-
esting that the radiation force method used by
Pohlman [1] in the first tissue absorption
studies is still perhaps the most reliable tech-
nique available. Acoustoel ectric receivers
show great promise for this kind of investiga-
tion [6] .
Almost all of the absorption studies have
used excised tissue samples. Very few in vivo
tissue absorption measurements have been made;
but those few are in rough agreement with
values for excised tissue. It was reported that
freshly excised liver had a much higher absorp-
tion than aged tissue samples at frequencies
near 1 MHz [7,8]. We have recently attempted
to duplicate these observations without success
[9]. Other aging studies have shown very little
change in absorption of tissues with time after
excision [10]. From the information available,
it appears reasonable to assume that the data
in figure 1 are valid for living tissues, but
we should realize that this is an assumption
which has not been thoroughly tested.
It is a surprising admission that after forty
years of study we do not know the temperature
dependence of the absorption coefficient for all
of the tissues of the body. The absorption co-
efficients for blood [11], brain [12], and myo-
cardial muscle [10] decrease slightly with tem-
perature in the range 0 to 40 °C. One study
reports a positive temperature coefficient for
bone between 0 to 60 °C [13], As discussed
below, mouse spinal cord is reported to have a
complex temperature dependence [14].
This discussion, of course, concerns longitu-
dinal ultrasonic waves in tissue. The propaga-
tion constants for shear waves in tissues are not
known in detail. However, recent rough measure-
ments indicate that transverse waves will not be
generated in any of the soft tissues of the body
with any current medical application of ultra-
sound [9,15].
2. Macromol ecul ar Absorption
The primary absorption of ultrasound in tissue
appears to take place at the macromol ecular level.
This was demonstrated directly by Pauly and Schwan
through a comparison in the absorption of sound
in liver tissue and a subcellular homogenate of
the tissue [16]. The near identity of the two
values suggests that tissue structure had little
influence on the absorption process.
A simpler biological system yields more de-
tailed information. The absorption coefficients
for packed red cells and for a solution of hemo-
globin at the same concentration in which it is
found in the erythrocyte {'^ 30 g/100 ml) are es-
sentially identical. This is shown as the upper
curves in figure 2. Now, if a 26 percent suspen-
sion of red cells in physiological saline is pre-
pared, the volumes of cells are the same as in
the packed preparation. Thus, the local concen-
tration and environment of the hemoglobin in the
cell remains the same, and hence the contribution
to the absorption from the hemoglobin should be
in direct proportion to amount of protein present
as shown by the dashed line in the figure 2 [17].
Saline contributes little to the absorption
30
2i
FREQUENCY (MHz)
Fig. 2. Absorption of sound in a suspension of
bovine erythrocyte in physiological saline
[17,18]. Dashed curve is the macromolecu-
lar contribution to the absorption of the
suspension (26% cells).
especially at low frequencies. Actual absorp-
tion measurements on such a suspension show that
the contributions of the hemoglobin solution in
the cell accounts for most, but not all, of the
observed loss. The excess absorption which
arises from the i nhomogeneous property of the
medium will be discussed later.
Since none of the smaller molecules or ions in
the red cell makes a significant contribution to
the absorption, it appears that the primary loss
comes from the presence of the protein itself.
Even the amino acid building blocks of hemo-
globin have much smaller specific absorptions
(i.e., absorption per gram solute) than the in-
tact protein as shown in figure 3. In fact, the
absorption of a 10 percent solution of amino
acids in the proportions found in hemoglobin is
barely distinguishable from water itself, where-
1 0 —\ 1 1 1 1 1 1 1 1 1
2 5 2 5 2 5
105 107 108 109
f (Hz)
Fig. 3. Absorption of sound in a 10 percent
hemoglobin solution at 25 °C [18]. Treat-
ment with the enzyme pronase fragmented
the protein yielding components with
molecular weight less than 10,000 and re-
sulting in a decreased absorption. A prep-
aration of amino acids in concentrations
equivalent to that in the hemoglobin solu-
tion give absorption values which are
hardly different than water alone.
as the absorption of a 10 percent solution of
hemoglobin is ten times that of water at 10 MHz
[18]. A pronase-treated hemoglobin solution
yielding subunits with molecular weight less than
10,000 showed a somewhat smaller specific absorp-
tion than that of the intact protein. A more
subtle dissection of hemoglobin using quanidine
hydrochloride produced specific absorptions which
were both larger and smaller than that of intact
hemoglobin as the fractionation proceeded [19].
The "whole much greater than the sum of its
parts" is not limited to hemoglobin or, for that
matter, to proteins. Figure 4 shows a very simi-
lar relationship among polysaccharides. The
absorption for dextran is much greater than for
simple sugars. It is interesting that the spe-
cific absorption of dextran and Ficoll molecules
of the same molecular weight are roughly equal.
Dextran is a long chain polymer whereas Ficoll
is globular. As a result, the macroscopic vis-
cosity of equivalent concentrations of the two
molecules is dramatically different. This is
just one of many observations that virtually
eliminate classical viscous processes from con-
sideration in tissue absorption.
^ 10-1
CM
4-
U
o
^ 2
9 MHz
sucrose
5 2
103
5 2
10"*
105
Molecular weight
Fig. 4.
Specific absorption of polysaccharides
vs_. molecular weight [18]. Data for
dextran are taken from reference [36].
Another polysaccharide, Ficoll, which
is nearly spherical and thus has a much
lower intrinsic viscosity than dextran,
has approximately the same specific
absorption as dextran.
A number of proteins, nucleic acids and poly-
saccharides have now been studied [2]. The spe-
cific absorption values for the macromolecules
are uniformly greater than those associated with
the small molecules and ions which make up tissue.
3. Macromolecular Interaction
The increase in specific absorption with mole-
cular complexity does not appear to stop at the
macromolecular level. Association among macro-
molecules in some cases results in increased
values for the absorption per molecule. This is
a vague loosely defined concept and is not based
on any specific, postulated mechanism. However,
there are a number of observations which seem to
fit this generic heading. The specific absorp-
tion of hemoglobin increases dramatically with
1Q2-
5-
" "^insulin Ficoll
o-dextran [36]
2-
lOi-
5-
10°-
• - hemoglobin solution
- erythrocyte suspension i.
30 MHz
10
— I —
20
30
40
50
60
70
g Hb/100 cm3
Fig. 5.
Absorption of hemoglobin as a function of
concentration at 25 °C [18]. For frequen-
cies greater than 10 MHz the absorption
coefficients of suspensions of red cells
are determined almost entirely by the
hemoglobin which they contain [11].
concentration as shown in figure 5 [18]. At a
concentration of 60 grams hemoglobin per 100 ml
solution, where the mean spacing of molecules is
very nearly equal to the molecular diameter, the
absorption per molecule is five times greater
than its value in a dilute solution. Whether
by introducing a new mechanism or modifying an
existing process, the interaction between macro-
molecules as the concentration increases causes
a marked increase in the specific absorption.
There are many other clear examples where
macromolecular interaction is associated with in-
creased absorption. The hemoglobin in rat
erythrocytes, when stored in sodium citrate at
4 °C for several days, becomes paracrystal 1 ine
(fig. 6) [20]. The specific absorption of para-
crystalline cells is at least twice that of
normal rat erythrocytes [ 18] . Treatment of red
cells with acrolein cross links the hemoglobin
and leads to specific absorption values which
are four times greater than the values for normal
erythrocytes as shown in figure 7 [21]. Other
examples may be found in figure 8. It is inter-
esting that the sediment from homogenized liver
has a specific absorption three times greater
than the "supernatant" from that preparation
[16]. Presumably the supernatant is a dilute
solution of macromolecules; whereas the sedi-
ment consists of highly organized subcellular
particles .
Perhaps these observations are no more than a
series of coincidences. However, the informa-
tion available today suggests that the trend to-
ward increasing absorption with increasing inter-
action extends beyond the macromolecular level
to interactions among macromolecules.
31
acrolein-fixed
erythrocytes
■ hypertonic L
erythrocytes J
• paracrystalllne ^
rat erythrocytes J-
erythrocyte '1
ghost membranej
yeast
normal rat
"erythrocytes
dilute albumin
normal bovine"L
erythrocytes J
-dilute hemoglobin
ficoll
pronase treated Hb
amino acids
5.0
muscle tissue
liver nuclei
3.0
liver tissue
liver sediment
2,0
1-0
0.5
liver homogenate
DNA, RNA
liver 'supernatant'
dextran
(MW > 30.000)
0.3
0.2
gelatin
amino acids
Fig. 6. Electron micrographic comparison of normal
and paracrystal 1 ine rat erythrocytes.
103-
asp
5-
rdB/cm]
Lg/cm^J
2-
102-
IQi
102
f (MHz)
Fig. 7.
Specific absorption for normal (o) and
fixed {o,x,/\) bovine erythrocytes in aque-
ous suspension at 25 °C. Acrolein concen-
trations in the original fixing solutions
were 0.3, 1.0 and 2 g/100 cm^. Cells were
washed free of excess acrolein before meas-
urement [21].
Fig. 8. Comparison of specific absorptions of
biological materials using dilute hemo-
globin as a reference. All data are for
10 MHz at 25 °C [18].
4. Relaxation Absorption
When stress-strain relationships for a medium
depend upon shifts in chemical or structural equi-
libria, it is possible that a sound wave propagated
in the medium will experience relaxation absorption.
Specifically, the absorption per wavelength is
[22,23].
1 +
(1)
where t is the relaxation time of the internal sys-
tem and c, cs, Cco are the phase velocity, and the
low and high frequency limits of the velocity re-
spectively. As implied by the equation, the magni-
tudes of the absorption and the dispersion in the
velocity are related. The relation has been tested
quantitatively for hemoglobin [11]. The results
are consistent with the hypothesis that a relaxa-
tion mechanism is responsible for the absorption
[24]. The first velocity dispersion measurements
reported for tissues support the hypothesis that
relaxation absorption is involved in these mater-
ials as well as in macromol ecular solutions. ^
2To a very rough approximation we can write [11]
aA * (f/c)(Ac/Af ). Kremkau [12] reports for
brain values of aA * 0.015, in agreement with
figure 1, and velocity dispersion of around
1 m/s/MHz near 1 MHz.
32
Frequency (MHz)
Fig. 9. Relaxation absorption. Log (aX) vs^ log
frequency is plotted for three temperatues
under assumptions discussed in the text.
In this example, the activation energy is
20 kcal/mol. The value for (aX)niax fot"
the illustration is arbitrary. With this
model, the (aX) curves simply shift to the
right, with log coq proportional to the in-
crement in temperature.
The frequency dependence of (aX) given by eq.
(1) is shown in figure 9. For certain simple re-
actions which may be perturbed by the sound wave,
we may write [25]
^ = . A e-E/f^T ■ (2)
where E is the activation energy of the dominant
rate process in the internal system, T the abso-
lute temperature, A and R are constants. If a
reference temperature Tq is chosen so that
T = To + AT, we can write
log wo * constant + -—^ • (3)
The general behavior predicted by figure 9 is seen
in a variety of simple systems i.e. the curves
of log (aX) vs . log f shift to the right in
increments proportional to the change in temper-
ature. This is illustrated with MnSO^ solutions
in figure 10 [26]. The temperature coefficient
of the absorption is negative if observations
are carried out somewhat below the relaxation
frequencies, while it is positive at high fre-
quencies. Figure 11 shows the temperature de-
pendence of (aX) for relaxation absorption with
activation energies of 20 and 50 cal/mol. As an
example, data for lecithin vesicles at 1 MHz are
shown [27].
Few biological materials can be characterized
by a single relaxation process. Rather, in most
cases there appear to be broad distributions of
relaxation frequencies, in the case of hemo-
globin, for example, extending from below 0.1
MHz to beyond 100 MHz (fig. 1). From our first
order model, we would anticipate that increasing
temperature would simply shift the curve of log
(aX) vs_. log f to higher frequencies. Where the
slope of this curve is slightly positive this
leads to small negative temperature coefficients
0.0
0.21 I I \ \ I \ I
0.2 0.4 0.6 1.0 2.0 4.0 6.0 10.0
Frequency (MHz)
Fig. 10. Absorption of sound in 0.5 molar MnSOi,
solutions in water. Dashed curves
represent the contribution of the low
frequency relaxation to the absorption
[26].
10-2i
Temperature (°C)
Fig. 11. Relaxation absorption. Log (aX) vs_.
temperature is plotted for activation
energies of 20 and 50 kcal/mol. As an
example, data (o) for lecithin vessicles
[27] are shown.
for the absorption. Brain [12], blood [11], and
heart muscle [10] (fig. 12) appear to have this
characteristic. The absorption per wavelength
in bone, as reported by Kishimoto [13] decreases
with frequency (fig. 13). As anticipated, this
leads to a positive temperature coefficient for
the absorption. In all of these materials, the
activation energies are of the order of a few
thousand kilocalories per mole.
33
The 1n v1vo data for mouse spinal cord [14]
are unique among the limited sample of tissues
for which temperature dependence of absorption
has been studied (fig. 14). In contrast with
other tissues which appear to have many relaxa-
tion frequencies, mouse spinal cord at 2 °C
shows a frequency dependence of (aA) which is
too strong to be explained even by a single re-
laxation frequency. In addition, the shape of
the log (aX) vs. log f curve changes with tempera-
ture. Because of these unusual properties, mouse
spinal cord appears to be a particular candidate
for further study. Since temperature studies
may provide general information about the proper-
ties of internal relaxing systems and in a sense
extend the range of frequencies observed, it
would be desirable to have this information for
a broad range of tissues.
1.0
Fig. 12.
2 5
Frequency (MHz)
10.0
Absorption of sound by myocardium [10].
The data illustrate the pairing of a
positive slope in log (aX) vs^. log fre-
quency with a negative temperature coef-
ficient for the absorption.
10-1
2-
10-
^s ingle
^v^relaxation
time
1.0-
0.14 ^ , 1
12 5 10
Frequency (MHz)
Fig. 13. Absorption of sound by bone [13]. In
this case the slope in log (aX) vs^. log
frequency is negative and the teniperature
coefficient of the absorption is positive
Frequency (MHz)
Fig. 14. Absorption of sound by mouse spinal cord
[14]. It appears the observations at
2 °C cannot be explained by relaxation
mechanisms.
Identification of the specific chemical or
structural reactions which are responsible for
relaxation absorption in tissues remains the
pristine challenge of this field of study. Al-
though some effort has been given to this prob-
lem, we must admit almost complete ignorance of
the nature of tissue relaxation mechanisms at a
meaningful, basic level.
5. Microscopically Inhomogeneous Materials
Although the absorption of sound in blood oc-
curs primarily at the macromolecular level, a
significant, measurable, contribution at fre-
quencies below 10 MHz arises from the presence of
intact red cells in suspension (fig. 2). Scat-
tering, viscous relative motion, and thermal
absorption have been considered as possible mech-
anisms for this excess loss [17,28-34]. Classi-
cal scattering is ruled out on qualitative and
quantitative grounds. Relative motion absorption
occurs when, because of density differences be-
tween suspended particle and suspending medium.
34
there is relative motion between the two phases
with a consequent viscous loss. A comparison of
observed, excess, non-protein absorption for sus-
pensions of red cells in physiological saline
with that predicted by theory for relative mo-
tion absorption is shown in figure 15 [17].
0.4 0.7
.0 2 4
Frequency (MHzI
20
Fig. 15. Comparison of non-hemoglobin absorption
in suspensions of red cells in saline
with that predicted by relative motion
absorption (solid curves). Points are
taken from the data in figure 2 and
reference [17].
Thermal absorption arises when thermal expansion
coefficients and specific heats of particles and
the suspending medium differ significantly thus
leading to irreversible heat flow between the two
phases. If extreme assumptions are made for the
thermal properties of blood cells, the predicted
thermal asborption is almost as large as that
predicted for viscous relative motion [33].
Definitive experiments to compare contributions
from the two mechanisms have not been attempted.
Present evidence, however, points to relative
motion as the dominant process. The suggestion
that relative motion may be important in the
solid tissues [31,34] remains to be tested.
Acknowledgments
The author wishes to acknowledge the contribu-
tion of Dr. Robert Weed for the use of the elec-
tron micrograph of a paracrystal 1 i ne red cell.
This review and much of the research reported
has been supported in part by U.S.P.H.S. Grant
GM09933.
References
[1] Pohlman, R., Ueber die Absorption des
Ultraschalls im menschlichen Gewebe und
ihre Abhaengi gkei t von der Frequenz,
Physik. Z. 40, 159-161 (1939).
[2] Johnston, R. L., Goss, S. A., Maynard , V.,
Brady, J. K. , Frizzell, L. A., O'Brien,
W. D., Jr., and Dunn, F., Elements of Tissue
Characterization. Part I. Ultrasonic Propa-
gation Properties (this publication, p. 19).
[3] Edmonds, P. D., Bauld, T. J., Ill, Dyro,
J. F. , and Hussey, M., Ultrasonic absorp-
tion of hemoglobin solutions, Biochim. et
Biophys. Acta 200, 174-177 (1969y
[4] Lemons, R. A. and Quate, C. F., Acoustic
microscopy: biomedical applications.
Science 188, 905-911 (1975).
[5] Marcus, P. W. and Carstensen, E. L.,
Problems with absorption measurements of
inhomogeneous solids, J. Acoust. Soc . Am.
58, 1134-1335 (1975).
[6] Busse, L. J., Miller, J. G., Yuhas, D. E. ,
Mimbs, J. W., Weiss, A. N., and Sobel ,
B. E., Phase Cancellation Effects: A
Source of Attenuation Artifact Eliminated
by a CdS Acoustoelectric Receiver in Ultra-
sound in Medicine, 0. White, ed. Vol. 3,
pp. 1519-1535 (Plenum Press, New York,
1977).
[7] Hueter, T. F., Visco-Elastic Losses in Tis-
sues in the Ultrasonic Range, Wright Air
Development Center. Wright-Patterson AFB,
Ohio, Tech. Rept. No. 57-706, ASTIA Doc. No.
AD142171 (August, 1968).
[8] Carstensen, E. L., The Mechanism of the
Absorption of Ultrasound in Biological Mate-
rials, IRE Trans, on Med. Electronics.
MET., 158-162 (1960).
[9] Frizzell, L. A. and Carstensen, E. L.,
Ultrasonic Heating of Tissues, University
of Rochester, Department of Electrical
Engineering Technical Report GM09933-20
(November, 1975).
[1^0] O'Donnell, M., Mimbs, J. W. , Sobel, B. E. ,
\ and Miller, J. G., Ultrasonic attenuation
of myocardial tissue: dependence on time
after excision and on temperature, J^.
Acoust. Soc. Am. (in press).
[11] Carstensen, E. L. and Schwan, H. P.,
Acoustic properties of hemoglobin solutions,
J. Acoust. Soc. Am. 31, 205-311 (1959).
[12] Kremkau, F. W., McGraw, C. P., and Barnes,
R. W. , Acoustic Properties of Normal and Ab-
normal Human Brain (this publication, p. 81)
[13] Kishimoto, T., Ultrasonic absorption in ,
bones, Acustica 8, 179-180 (1959).
[14] Dunn, F. and Brady, J. K., Ultrasonic ab-
sorption in biological media, Biofizika
18, 1063-1066 (1973); Biophysics 18, 1128-
1132 (1974).
[15] Frizzell, L. A. and Carstensen, E. L.,
Shear properties of mammalian tissues at
low megahertz frequencies, J. Acoust. Soc.
Am. 60_, 1409-1411 (1976).
[16] Pauly, H. and Schwan, H. P., Mechanism of
absorption of ultrasound in liver tissue,
J. Acoust. Soc. Am. 50, 692-699 (1971).
35
[17] Carstensen, E. L. and Schwan, H. P.,
Absorption of sound arising from the pres-
ence of intact cells in blood, J. Acoust.
Soc. Am. 31, 185-189 (1959).
[18] Kremkau, F. W. and Carstensen, E. L.,
Macromolecular Interaction in Sound Absorp-
tion, Proc. of Workshop on Interaction of
Ultrasound and Biological Tissues, Battelle
Seattle Research Center, Food and Drug
Administration, U.S. Dept. of Health Educa-
tion and Welfare, Rockville, Maryland
(1972), pp. 37-42.
[19] O'Brien, W. D. and Dunn, F., Ultrasonic
examination of the hemoglobin dissociation
process in aqueous solutions of guanidine
hydrochloride, J. Acoust. Soc. Am. 50,
1213-1215 (1971'r
[20] Ponder, E. , The paracrystall i ne state of
the rat red cell, J. Gen. Physiol. 29,
89-102 (1945).
[21] Kremkau, F. W. and Carstensen, E. L.,
Macromolecular interaction in the absorp-
tion of ultrasound in fixed erythrocytes,
J. Acoust. Soc. Am. 53, 1448-1451 (1973).
[22] Kremkau, F. W. and Carstensen, E. L.,
Macromolecular Interaction in the Absorp-
tion of Ultrasound in Biological Material,
University of Rochester, Department of
Electrical Engineering Tech. Report, No.
GM09933-18 (April, 1972).
[23] Markham, J. J., Beyer, R. T., and Lindsay,
R. B., Absorption of sound in fluids.
Rev, of Mod. Phys. 22, 353-411 (1951).
[24] Eigen, M. and deMaeyer, L., Relaxation
Methods, in Technique of Organic Chemistry,
A. Weissberger, ed. , Vol . 8, Part II, pp.
895-1054 (Wiley, New York, 1963).
[25] Stuehr, J. and Yeager, E., The Propagation
of Ultrasonic Waves in Electrolytic Solu-
tions, in Physical Acoustics, W. P. Mason,
ed.. Vol. 2, Part A, pp. 351-462 (Academic
Press, New York, 1965).
[26] Carstensen, E. L., Relaxation processes
in aqueous solution of MnSO^ and CoSO^,
J. Acoust. Soc. Am. 26, 862-864 (1954).
[27] Eggers, F. and Funck, T., Ultrasonic re-
laxation spectroscopy in liquids, Natur-
wissenschaften 63, 280-285 (1976).
[28] Epstein, P. S. , On the Absorption of Sound
Waves in Suspensions and Emulsions, Theo-
dore von Karman Anniversary Volume, Cali-
fornia Institute of Technology, Pasadena,
California, 1941.
[29] Epstein, P. S. and Carhart, R. R. , The
absorption of sound in suspensions and
emulsions. I. Water fog in air, J. Acoust.
Soc. Am. 25, 553-565 (1953)
[30] Urick, R. J., The absorption of sound in
suspensions of irregular particles, J_.
Acoust. Soc. Am. 20, 283-289 (1948).
[31] Fry, W. J., Mechanism of acoustic absorp-
tion in tissue, J. Acoust. Soc. Am. 24,
412-415 (1952).
[32] Allegra, J. B. and Hawley, S. A., Attenua-
tion of sound in suspensions and emulsions:
theory and experiments, J. Acoust. Soc.
Am. 51, 1545-1564 (1972).
[33] Ahuja, A. J., Acoustical properties of
blood: a look at the basic assumptions,
Med. Phys. 1, 311 (1974).
[34] O'Donnell, M. and Miller, J. G., Mecha-
nisms of Ultrasonic Attenuation in Soft
Tissue (this publication, p. 37).
[35] Kessler, L. W. , VHF ultrasonic attenuation
in mammalian tissue, J. Acoust. Soc. Am.
53, 1759-1760 (1973).
[36] Hawley, S. A. and Dunn, F., Ultrasonic
Absorption in Aqueous Solutions of Dextran,
J. Chem. Phys. 50, 3523-3526 (1969).
36
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed., National Bureau
of Standards, Spec. Publ . 525 (U.S. Government Printing Office, Washington, D.C., 1979).
MECHANISMS OF ULTRASONIC ATTENUATION IN SOFT TISSUE
M. O'Donnell and J. G. Miller
Laboratory for Ultrasonics
Department of Physics
Washington University
St. Louis, Missouri 63130, U.S.A.
Ultrasonic loss mechanisms which arise from the microscopically i nhomogeneous nature
of soft tissue are investigated over the frequency range 1 to 10 MHz. Contribu-
tions to the ultrasonic attenuation due to viscous relative motion losses and thermal
losses are shown to exhibit an approximately linear dependence on frequency. Numerical
estimates of the attenuation arising from these mechanisms are compared with the re-
sults of experiments for a representative substance in each of three attenuation cate-
gories: low (blood), medium (heart), and high (skin). Inhomogeneity losses can account
for 60 percent of the attenuation observed in heart and skin, and thus may contribute
a non- negl igi bl e fraction to the attenuation observed in most soft tissue.
Key words: Inhomogenei ties ; inhomogenity thermal losses; mechanisms; viscous relative
motion .
In efforts to account for the ultrasonic at-
tenuation exhibited by soft tissue, two types
of mechanisms have been considered: (1) struc-
tural or chemical relaxation of the macro-
molecular constituents [1-3]^, and (2) proc-
esses arising from the i nhomogeneous nature of
tissue, including scattering, viscous relative
motion, and thermal losses [1,3-8]. In this
paper we estimate the fraction of the total at-
tenuation observed in the 1 to 10 MHz frequency
range which is attributable to processes arising
solely from the microscopically i nhomogeneous
nature of soft tissue, as opposed to that frac-
tion presumed to arise from macromol ecu! ar re-
laxation processes.
We consider a model in which a longitudinal
ultrasonic wave propagates through tissue which
is viewed as a suspension of scatterers in a
liquid medium (cytoplasm). For the sake of
simplicity all scatterers are considered to be
spherical in shape. At an interface between
the medium and a scatterer, longitudinal, vis-
cous, and thermal waves may be excited in the
suspending medium due to the discontinuity of
acoustic and thermal properties at the surface
1 of the scatterer. Energy that is coupled into
!] these waves is lost from the incident ultra-
i sonic beam and thus contributes to the total
1 attenuation that is observed.
For scatterers small compared to the wave-
[j length of the incident wave the reradiated
1, longitudinal wave corresponds to Rayleigh
j scattering. The effect contributes a term to
■ the attenuation coefficient that varies as
the fourth power of frequency and the third
^Figures in brackets indicate literature
references at the end of this paper.
power of the scatterer radius for a fixed
volume concentration of scatterers. The
magnitude of Rayleigh scattering depends upon
the square of the difference in adiabatic
compressibilities of the scatterer and the
suspending medium and upon the square of the
difference in densities. The numerical value
of the contribution due to Rayleigh scatter-
ing is several orders of magnitude lower than
the observed attenuation coefficient of soft
tissue in the frequency range 1 to 10 MHz [1,2].
Details of the scattering process for scat-
terers of dimensions which are not small com-
pared to the ultrasonic wavelength are less well
understood. (Specular reflections from objects
much larger in size than the ultrasonic wave-
length are explicitly excluded here.) The
general features of the scattering problem in-
dicate that the magnitude of scattering events
depends primarily on the square of the differ-
ences in densities and compressibilities of the
scatterer and the suspending medium. In soft
tissue, scatterers of dimensions comparable to
ultrasonic wavelengths in the 1 to 10 MHz range
(i.e., scatterers of sizes ranging from 0.05
mm to 1 mm) exhibit densities and compressi-
bilities yery close to those of the suspending
medium. Thus losses arising from these scat-
tering events are expected to represent small
contributions to the attenuation coefficient.
Such scattering events, however, may be of
central importance in scattering and reflec-
tion (i.e., 180° backscattering) experiments.
Viscous drag losses arise from the genera-
tion of a highly damped viscous wave in the
suspending medium. The scatterers attempt to
mirror the motion of the suspending medium to
all extent determined by a function which de-
pends upon the relative densities of the scat-
37
terers and the medium, the ultrasonic fre-
quency, and the viscosity of the medium [4,7].
For fixed frequency, viscosity, and volume
concentration of scatterers, the contribution
of viscous drag losses to the attenuation co-
efficient is predicted to rise, reach a maxi-
mum, and subsequently fall as a function of
scatterer size. A similar rise, plateau, and
fall is predicted if viscous relative motion
losses are plotted as a function of frequency
for fixed scatterer size. Contributions to the
attenuation of ultrasound in tissue by viscous
relative motion losses were considered by Fry
[6], Carstensen and Schwan [1], and Kremkau,
Carstensen, and Aldridge [3].
An additional contribution to the ultrasonic
attenuation arises from the generation of a
highly damped thermal wave at the surface of a
scatterer. (Strictly speaking, the viscous drag
and thermal effects are coupled, but the strength
of this coupling is sufficiently small that it
can be ignored [4].) The magnitude of the losses
associated with the generation of the thermal
wave depends upon frequency and upon the relative
heat capacities, coefficients of thermal expan-
sion, and thermal conductivities of the scat-
terers and suspending medium. At low frequencies
the system is in an isothermal limit in which
energy is reversibly exchanged between the car-
rier medium and a scatterer, which closely follow
each other in temperature ("isothermal"), so that
relatively little loss of energy occurs. At high
frequencies the system is in an adiabatic limit
in which temperature changes in the suspending
medium occur so rapidly that they cannot be com-
municated to a scatterer. Thus little heat
transfer occurs ("adiabatic") and relatively
little energy is dissipated. Maximum energy is
dissipated in the intermediate range of fre-
quencies where the thermal response of the scat-
terers to the incident compressi onal ultrasonic
wave lags sufficiently behind that of the sus-
pending medium that substantial energy is dis-
sipated. Thermal losses in i nhomogeneous media
have been considered by Epstein and Carhart [4],
Allegra and Hawley [7], Kremkau, Carstensen, and
Aldridge [3], and Ahuja [8].
In order to estimate the contribution to the
observed attenuation arising from viscous and
thermal effects, it is necessary to identify the
relevant scatterers in biological specimens. In
principle, all inhomogeneities contribute to the
observed attenuation at any frequency. The
physical arguments presented above, however,
suggest that for a fixed range of frequencies
only scatterers exhibiting specific physical
properties that define the relatively narrow
ranges over which either viscous or thermal
losses are near maximum need be considered.
Numerical computations discussed below indicate
that for the range of scatterer dimensions
which results in significant viscous relative
motion losses or thermal losses, contributions
to the attenuation coefficient exhibit an ap-
proximately linear dependence on frequency,
i.e., log (attenuation) versus log (frequency)
exhibits a slope of 0.8 to 1.2 over the fre-
quency range 1 to 10 MHz. Therefore, the slope
of the attenuation versus frequency curve over
the range of 1 to 10 MHz (hereafter referred to
as the slope of the attenuation) can serve as a
useful index of ultrasonic properties. This
conclusion, arrived at from theoretical con-
siderations, is consistent with the results of
experiments, which indicate an approximately
linear dependence on frequency for the ultra-
sonic attenuation coefficient of soft tissue.
The slope of the attenuation represents a de-
sirable index since in transmission experiments
it is less susceptible to errors associated
with impedance discontinuities and may be less
susceptible to errors resulting from phase
cancellation effects than is the apparent at-
tenuation coefficient.
LO
U.8
/ \ THERMAL ATTENUATION
'e
/ \ Con^ = 17%
b
0.6
/ \ Pscatt = 1 4lg/cm3
Its of
0.4
c
0.2
SLOPE
0.0
1 1 . 1
0.01 01 I 10 100
SCATTERER DIAMETER ( /xm )
Fig. 1. Slope of the attenuation versus frequency
curve over the range 1 to 10 MHz arising
from thermal losses plotted as a function
of scatterer diameter.
In figure 1 the contribution to the slope of
the attenuation arising from thermal losses is
plotted as a function of scatterer size. The
physical properties of the scatterers and the
suspending medium used in calculating the re-
sults shown in figure 1 appear in table 1 which
is discussed below. As illustrated in figure 1,
contributions from thermal losses are largest
when the thermal wavelength Xi of the suspending
medium approximately equals 2 ir times the radius
of the scatterer. (A^ is on the order of 0.4 to
1.3 m in the frequency range 1 to 10 MHz.)
The slope of the attenuation which results
from the coupling of energy into a viscous wave
is plotted as a function of scatterer size in
figure 2. As illustrated, viscous relative mo-
tion losses are largest when the viscous wave-
length Ay of the suspending medium approximate-
ly equals 2 ir times the radius of the scatterers.
(Ay is the order of 1.8 to 5.5 vim in the 1 to
10 MHz frequency range. )
Our calculations indicate that viscous rela-
tive motion losses dominate thermal losses in
the 1 to 10 MHz frequency range for a wide
range of scatterer properties consistent with
the microscopic anatomy of soft tissue. We
thus focus on viscous relative motion losses in
an attempt to identify specific physical proper-
ties that must be exhibited by the relevant
scatterers. The numerical magnitude of the
viscous relative motion term is approximately
proportional to (p'/p - 1)^, where p' is the
density of the scatterer and p is the density
of the suspending medium. Therefore, signifi-
cant contributions to the observed attenuation
result only from those scatterers which (1) ex-
hibit sizes that maximize viscous relative mo-
38
75
X 60
45
30
/~\ VISCOUS ATTENUATION
/ \ Con^, = 17%
/ \ '^suspending = 2.5cp
/ \ medium
-
-
/ \ ^scQtt = I.41g/cm3
\ ■ - '
- , 1 1
111 1 1
0.01 0.1 I 10
SCATTERER DIAMETER (/xm)
100
Fig. 2. Slope of the attenuation (1 to 10 MHz)
arising from viscous relative motion
losses plotted as a function of scatterer
diameter.
tion losses, (2) have densities significantly
different from that of the suspending medium,
and (3) are present in sufficient concentra-
tions. Aggregates of structural protein such
as fibrils of collagen or muscle myofibrils ap-
pear to be the principal constituents of soft
tissue which possess these properties. (Dunn
and his colleagues have identified collagen as
a significant determinant of the ultrasonic
properties of soft tissue, in part because of
its large compressional modulus and ubiquity
[9,10].)
Limiting our attenuation to scatterers ex-
hibiting the requisite physical properties, we
estimate the contribution to the observed at-
tenuation arising from microscopic inhomo-
geneities. Specific numerical values for the
properties of the scatterers and the suspend-
ing medium which were used in our calculations
are presented in table 1. An attempt has been
made to estimate the contribution to the ex-
perimentally observed attenuation arising only
from viscous and thermal effects for a repre-
sentative substance in each of three attenua-
tion categories: low (blood), medium (heart).
and high (skin). Results of these calculations
are summarized in table 2. Losses associated
with viscous relative motion are seen to domi-
nate thermal losses in the 1 to 10 MHz range,
a feature which appears to be preserved over a
wide range of scatterer properties.
The quantitative results presented in table
2 indicate that losses arising solely from
microscopic inhomogeneities appear to account
for a significant fraction ('v 60 percent) of
the total observed attenuation of heart and
skin. These results suggest that i nhomogeneity
losses may contribute a non-negligible fraction
to the attenuation of most soft tissue. Al-
though the exact magnitude of these losses is
certain to be altered by the use of more real-
istic assumptions (e^. , non-spherical scat-
terers and improved estimates of the values of
physical properties specified in table 1), the
general conclusions of this study would not be
altered if the illustrative numerical results
presented in table 2 were in error by 30 percent.
Table 2. Slope of attentuation coeff icient-vs-frequency
(cm-iMHz-i
Tissue Viscous Thermal Theoretical Experimental Percent
value^ value accounted forC
Blood
(40% Hct)
Heart 0.042
Skin 0.101
0.003 0.000
0.001
0.001
0.003
0.043
0.102
0.03ld
10
0.0726 60
0.17of 60
Numerical results presented to 3 significant digits for
j^i 1 1 ustrati ve purposes only.
^Sum of contributions from columns 2 and 3.
Percent of experimentally observed values (column 5) accounted
for by sum of contributions (column 4) arising from losses due
.to microscopic inhomogeneities.
Carstensen, E. L., Li, K. , and Schwan, H. P., J. Acoust. Soc.
Am. 25, 286-289 (1953).
^O'Donnel, H., Mimbs, J. W. , Sobel , B. E., and Miller J. G.,
Ultrasonic Attenuation in Normal and Ischemic Myocardium
^(this conference).
Dussik, K. T., Kyriazidov, M., Fritch, 0. J., and Sear, R. S.,
Am. J. Rhys. Med. 37, 160 (1958).
Table 1. Acoustic and thermal properties of tissue scatterers (T = 20 'C).
Tissue
Scatterers
Size range^
Density
Vol ume
Thermal
Coefficient
Speci f ic
concentration conductivity
of thermal
heat
expansion
% of total
10-5
(pm)
(g/cm3)
vol ume
cal/s-cm-^C
10-" deg-i
cal /g-°C
Blood
Red cells
4.6 - 5. 6b
1 .09b
40b
1.1b
1.2c
0.8b
(40% Hct)
1 .32^
Heart
Muscl e
1,0 - 2.0^
18d
0.4c
1 .2C
0.4c
myof i bri 1 s
1 .4lf
Skin
Ccl lagen
0.7 - 1.56
259
0.4c
1 .2=
0.4c
f ibri 1 s
1 .03b
Suspendi ng^
1 .5b
1.41
1 .Ob
— medium
.Uniform distribution of scatterers with diameters expressed in micrometers.
°Ahuja, A. S., Med. Phj^. 1, 311-316 (1974).
^Estimated .
Bailey, Kenneth, Proteins , H. Neurath and Kenneth Bailey, eds., Chapt. 4 (Academic Press,
New York, 1954).
.pGross, Jerome, Scientific American, 120-130 (May 1961).
Pomeroy, C. D. and Milton, R. J., J. Soc. Leather Trades Chem. 35, 360 (1951).
^Chupvil , .M. , Physiology of Connective Tissue (Butterworth, London, 1967).
.Viscosity (1 to 10 MHz) estimated to be 2.5 centipoise.
Epstein, P. S. and Carhart, R. C, J. Acoust. Soc. Am. 25, 553 ( 1953).
39
Acknowl edgment
This work was supported in part by grants
HL19537, HL17646, and HL07081 from the National
Institutes of Health.
Pranoat Suntharothok-Priesmeyer was respon-
sible for production of the text and illustra-
tions.
References
[1] Carstensen, E. L. and Schwan, H. P.,
Absorption of sound arising from the
presence of intact cells in blood, J_.
Acoust. Soc. Am. 31_, 185 (1959).
[2] Pauly, H. and Schwan, H. P., Mechanism of
absorption of ultrasound in liver tissue,
J. Acoust. Soc. Am. 50, 692 (1971).
[3] Kremkau, F. W. , Carstensen, E. L., and
Aldridge, W. G., Macromolecular inter-
actions in the absorption of ultrasound
in fixed erythrocytes, J. Acoust. Soc.
Am. 53, 1448 (1973).
[4] Epstein, P. S. and Carhart, R. C, The
absorption of sound in suspensions and
emulsions. I. Water fog in air, J_.
Acoust. Soc. Am. 25., 553 (1953).
[5] Urick, R. J., The absorption of sound in
suspensions of irregular particles, J_.
Acoust. Soc. Am. 20, 283 (1948).
[6] Fry, W. J., Mechanism of acoustic absorp-
tion in tissue, J. Acoust. Soc. Am. 24
412 (1952).
[7] Allegra, J. R. and Hawley, S. A., Attenua-
tion of sound in suspensions and emulsions
theory and experiments, J. Acoust. Soc. Am
51, 1545 (1972).
[8] Ahuja, A. J., Acoustical properties of
blood: a look at the basic assumptions,
Med. Phys. U 311 (1974).
[9] Fields, S. and Dunn, F., Correlation of
echographic visualization of tissue with
biological composition and physiological
state, J. Acoust. Soc. Am. 54, 809 (1973).
[10] O'Brien, W. D. Jr., The Role of Collagen
in Determining Ultrasonic Propagation
Properties in Tissue, in Acoustical Holo-
graphy, L. W. Kessler, ed. , Vol . 7 (Plenum
Press, New York, 1976).
CHAPTER 3
ATTENUATION AND VELOCITY II:
METHODOLOGY AND MEASUREMENTS
41
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer , ed. , National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
ELEMENTS OF TISSUE CHARACTERIZATION
Part II. Ultrasonic Propagation Parameter Measurements
S. A. Goss, R. L. Johnston, V. Maynard, L. Nider,
L. A. Frizzell, W. D. O'Brien, Jr., and F. Dunn
Bioacoustics Research Laboratory
University of Illinois
Urbana, Illinois 61801, U.S.A.
Methods employed at the Bioacoustics Research Laboratory of the University of
Illinois for the determination of ultrasonic propagation properties of biological
media are described, with attention devoted to attenuation, absorption and veloci-
ty measurements of both longitudinal and shear ultrasonic waves. These include
systems specifically for the ultrasonic characterization of soft tissues and for
solutions of biologically significant macromolecules. Each method is presented
in terms of theory, limitations, applicability, and possible sources of error.
Important new techniques from other laboratories are also discussed.
Key words: Ultrasonic absorption; ultrasonic attenuation; ultrasonic instrumenta-
tion; ultrasonic measurements; ultrasonic spectroscopy; ultrasonic
tissue characterization; ultrasonic tissue parameters; ultrasonic
tissue signature; ultrasonic velocity.
1. Introduction
A number of measuring techniques have been
developed, or are otherwise employed, at the
Bioacoustics Research Laboratory of the Uni-
versity of Illinois in the research efforts as-
sociated with the propagation characteristics of
ultrasound in biological tissue, and toward a
basic understanding of the mechanism(s) re-
sponsible for the acoustic properties exhibited
by those tissues. These include specialized
systems for biological liquids and for soft
tissues which are capable of measuring attenua-
tion, absorption, and velocity of longitudinal
ultrasonic waves as functions of frequency, tem-
perature and, where appropriate, pH and ambient
pressure. In addition, a system is presently
being developed to measure the shear acoustical
properties of biological materials as a function
of frequency and temperature. These techniques,
thirteen in all, are described in this paper in
terms of parameters measured, theory, applica-
tion, precision/accuracy, frequency, and tempera-
ture range, so that each technique may be as-
sessed separately.
2. Attenuation and Velocity
Measurements in Soft Tissue
Attenuation of longitudinal waves in soft tis-
sue specimens are determined using the radiation
force, pulse transmission, and standing wave
techniques. Velocity measurements in soft tis-
sues can be made by observing the time of flight
of an acoustic pulse through a known path length
of a tissue sample, and by acoustic interferometry.
A. Attenuation: Radiation Force Method
The phenomenon of radiation force provides a
primary method for the measurement of the second
order quantities of intensity and power [l-4]i.
Radiation force is a direct result of energy
transport by the sound wave, and is equal to the
time rate of change of momentum of the wave.
Thus a continuous wave incident on a reflecting
or absorbing object will produce a time inde-
pendent force on that object equal to, and in
the direction of, the time rate of change of
momentum. Consequently, a sensitive balance can
be employed to measure the radiation force exert-
ed by an ultrasonic beam incident on a suspended
target. It then follows that force, measured as
a change in tension of the suspending wire, on a
perfectly absorbing target intercepting a verti-
cally directed sound beam is
where F is the measured change in tension of the
suspending wire, P is the time average power in-
tercepted by the target, and c is the velocity
of propagation. Thus, in water, 1 milliwatt of
incident power will exert a force equivalent to a
weight of 67 pg, which is measured as an apparent
target weight change. Similarly, a perfectly
reflecting target intercepting a vertically
directed sound beam will produce a measurable
change in suspension tension, given as
Figures in brackets indicate literature
references at the end of this paper.
43
F=2Pcos29
c
where e is the angle between the beam axis and the
normal to the target surface. Consequently in
water, when the target is inclined 45° to the ver-
tical, the sensitivity of the balance is again 67
pg/mW of intercepted power. Thus, a phase insensi-
tive, frequency independent technique for power and
intensity attenuation measurements becomes available
by interposing a tissue sample between the sound
source and the sound intercepting target, by deter-
mining that power lost or redirected with the tissue
sample in place compared with measured power with
the tissue absent [2,5,6]. Care must be taken to
insure that the system's sensitivity is actually that
that predicted for such idealized cases. The atten-
uation present in the liquid media (water or physio-
logical saline) will inevitably cause acoustic
streaming which will lead to systematic errors in
the power estimation. The streaming effect can be
minimized by positioning an acoustically transparent
barrier of, for example, stretched polyethylene,
directly in front of the target.
A second possible source of error is due to the
inherent insensitivity of the measurement system
to horizontal forces. For maximum accuracy, the
main lobe of the sound beam should be directed
precisely vertically. Because the beam may con-
tain components of momentum in the plane perpen-
dicular to the main lobe, the radiation force
balance method does not respond to all the energy
emitted by the source. As the temperature of the
target increases due to the fact that energy is
being deposited in it, thermal expansion of the
target can introduce an error by virtue of in-
creasing its buoyancy and resulting in an apparent
target weight change. The ultimate accuracy of
the radiation force balance method is limited by
noise associated with mechanical vibrations and
atmospheric perturbations, which can be reduced by
proper design considerations. However, in the
presence of pressure gradients, the target is
acted on by forces proportional to the target
volume, making a small target advantageous.
B. Attenuation: Pulse Transmission Method
The pulse transmission technique is applicable
to measurements of sound pressure attenuation in
various types of biological specimens [7,8]. The
instrumentation consists of a transmitting ultra-
sonic transducer immersed in a bath (usually
physiological saline) which serves as the acoustic
coupling medium and aligned axially with either a
reflecting surface or a receiving transducer. The
associated electronic instrumentation provides an
RF pulse to the transmitting crystal and also trig-
gers the sweep of an oscilloscope a variable, but
selected, length of time after the initiation of
the transmitting pulse. An oscilloscope serves
to display the amplified received signal. Measure-
ment of sound attenuation can be performed as fol-
lows. First, with no sample in the ultrasonic
path, the gain of the receiving amplifier is ad-
justed to give a signal display of a predetermined
amplitude. Then, with the specimen in place, the
gain of the system is adjusted, either by increas-
ing the receiving amplifier gain or by removing
electrical attenuation in the signal path, to com-
pensate for the loss of acoustic energy in the
sample to bring the display trace to the same
preselected amplitude. The attenuation measure-
ment is repeated for other samples of the same
material, but of varying thicknesses. The mea-
sured attenuation values are then plotted versus
sample thickness, and the slope of the resulting
straight line yields the attenuation per unit
length of the specimen.
C. Velocity: Pulse Transit Time Method
Velocity measurements in tissue can be made
simply by measuring the time of flight of an
acoustic pulse over a known path length of the
specimen [9,10]. Differential techniques should
be used to minimize uncertainties in path length
and t^'me delay measurements, such as the uncertain-
ty of the actual location of the active face of
the transducer. If commercial transducers are
employed, additional allowances must be made for
the quarter-wave impedance matching layer on their
surface. Errors may be encountered in time of
flight measurements of the acoustic signal in in-
homogeneous media, such as tissues. Since a short
duration acoustic pulse contains a broad spectrum
of components, the frequency dependent effects of
velocity dispersion, attenuation, and multiple
phase shifts at tissue interfaces, can distort
and delay the signal [11]. Techniques which
rely on the detection of the leading edge of the
received acoustic pulse for timing are prone to
the greatest inaccuracies under such distortions.
D. Velocity: Acoustic Interferometric
Method
Nearly an order of magnitude improvement in ac-
curacy can be realized by the use of acoustic in-
terferometric techniques employing CW excitation
[12,13]. By sandwiching the tissue sample between
transmitting and receiving transducers, or between
a transducer and a reflecting target, a standing
wave is created in the tissue sample. It has been
shown that tissue may be distorted in this manner
up to 25 percent without affecting the measured
velocity [12]. By monitoring the maxima and
minima of the received acoustic signal as the path
length is changed, the standing wave pattern and
the wavelength of sound in the tissue can be mea-
sured. This information combined with the fre-
quency employed for the measurement yields the
velocity of sound in the sample. Alternatively,
the path length may be held fixed and the frequen-
cy of excitation changed. The change in frequency
necessary to move from one mode in the standing-
wave pattern to the next is directly proportional
to velocity. The accuracy of such interferometric
techniques, under tight temperature control, can
reach about 0.1 percent.
E. Attenuation and Velocity: Standing
Wave Method
While the measuring techniques described above
may be applied to the measurement of the ultra-
sonic absorption and velocity of most biological
tissues, some specimens require specialized
techniques .
Lung, by virtue of its extraordinarily high
attenuation, presents some problems in determina-
tion of its ultrasonic propagation properties.
The following method has been developed and yields
results in agreement with other similar ones [8].
44
Briefly, the lung is suspended in the sound field
between the sound source and an absorption chamber,
to eliminate standing waves beyond the lung. The
transient thermoelectric probe [14] provides a
convenient acoustic detector to investigate (A)
the acoustic field between the specimen and the
source, to determine the axial standing wave pat-
tern, and (B) the acoustic field between the speci-
men and the absorption chamber, to determine the
wave amplitude transmitted beyond the lung speci-
men 15 (see fig. 1). Probing the field in A
yields the fraction of incident energy reflected
at the lung-saline interface. Assuming that in-
finitesimal wave acoustics prevails and that the
attenuation in the lung is sufficiently great that
multiple reflections within the specimen need not
be considered, the speed of sound in the lung can
be obtained from
!pv)i
ung
(pv)saline
SWR
(3)
where the p's are the known densities and SWR is the
observed standing wave ratio (between the source and
specimen) and is greater than unity. The attenua-
tion coefficient per unit path length is determined
from a knowledge of the energy reflected at the two
lung-saline interfaces, the thickness of the sample,
and the acoustic intensity detected by the probe in
B in accordance with the relation
Ine
-2al
(4)
where Iq and Ij are, respectively, the acoustic in-
tensities at the lung-saline interface nearest to
and farthest from the source, 1 is the thickness
of the sample, and a is the attenuation coefficient.
3.
Absorption Measurements
in Soft Tissue
The transient thermoelectric method is well
suited for determining acoustic absorption in
small volumes of highly absorbing liquid or tis-
sue ir[ vivo as well as i_n vitro, and may be the
only method, applicable to tissues, that mea-
sures directly absorption rather than attenua-
tion. As such it is invaluable in determining
the portion of attenuation due to absorption vis-
a-vis that portion due to scattering effects
L16,17]. The method requires that a thermocouple
junction of small diameter relative to the wave-
length be implanted in the sample. The thermo-
couple and surrounding sample are then exposed to
a plane traveling wave ultrasonic field having a
temporally rectangular envelope of known intensity.
The typical thermoelectric emf response to a 1
second ultrasonic pulse has an initial fast rise
which results from conversion of acoustic energy
into heat by the viscous forces acting between
the thermocouple wire and the sample. This por-
tion of the response approaches equilibrium very
quickly with a magnitude that depends mainly upon
the thermocouple wire radius, the viscous proper-
ties of the sample medium, the sound intensity,
and the frequency. The fast rise is followed by
a relatively linear rise (in the absence of
thermal conduction processes) that is a result of
absorption of the ultrasound in the surrounding
medium. From a determination of the slope of the
1 inear portion of the thermoelectric emf response
as a function of time, the absorption coefficient
can be calculated using the following relation,
CnK
p'-p
21
/dT\
^dt/n
(5)
where a is the absorption coefficient (cm"i), p
is the density (g/cm^), Cp is the specific heat at
constant pressure (cal/°C g), K is the mechanical
equivalent of heat (4.18 J/cal), I is the acoustic
intensity (W/cm^), and (dT/dt)o is the initial
time rate of change of temperature due to absorp-
tion in the medium (°C/s). Effects limiting the
i
pC — rubber diaphragm
position of probe
for determination
of absorbed energy
\
position of probe
for determination
of reflected energy
(A)
castor oil
absorption
chamber
coupl ing
medium
position of excised
lung capsule
sound
tank
Fig. 1,
Schematic diagram of the experimental arrangement for attenuation and velocity
measurements in lung using the standing wave method [15].
45
accuracy of the method include any changes of the
thermal properties of the tissue and/or the absorp-
tion itself with temperature. Provided the tem-
perature rise during measurement is on the order
of one degree or less these effects will be
minimal. Dunn [18] has indicated that the total
uncertainty in the determination of (dT/dt)n is of
the order of 5 to 10 percent. In application of the
transient thermoelectric method the initial value
of a is determined by using the ultrasonic in-
tensity that would be present at the site of the
junction were the sample absent. The measured
depth of the thermocouple and the initial value of
a are then used to calculate intensity at the site
of the junction by correcting for absorption using
an iterative method [18] until the value of o con-
verges (note that if the value of a and/or the
depth of the thermocouple are too large, the value
of a will not converge).
This technique has been used at frequencies as
low as 0.26 MHz [19] and as high as 7 MHz [20] in
tissues, but modifications have been used to 2 GHz
in fluid media [21]. The lower frequency limit is
determined by the value of the absorption, since
too low an a yields too low a temperature rise to
be detected accurately without having to increase
the exposing intensity to unacceptable levels.
Also, at low frequencies and absorptions the ini-
tial viscous heating portion of the transient
thermal emf may become a major portion of the
signal, making the subtraction process a relatively
erroneous one. At higher frequencies the limita-
tions on accuracy of the method, as described for
tissues, are imposed by (1) availability of broad
plane wave ultrasonic fields, (2) the size of the
thermocouple relative to the wavelength so that
the field is affected by scattering from the wire
and thermal conduction along it, and ultimately
by (3) nonconvergence in the iterative method of
depth correction due to high absorption. The lat-
ter two limiting effects have not yet been trouble-
some in the region of application. The method has
been found to be very useful and easily applied to
most tissues over the frequency range from 0.5 to
4.0 MHz.
4. Absorption and Velocity Measurement
in Biological Liquids
The primary mechanism(s) for the absorption of
sound in biological material seems to occur at
the macromolecul ar level. Carstensen, Li, and
Schwan [22] discovered that the acoustical proper-
ties of blood are determined largely by the
protein content, and that the absorption coeffi-
cient is directly proportional to the protein
concentration, whether in solution or within the
cell. Pauly and Schwan [23] have demonstrated
that nearly two-thirds of the ultrasonic absorp-
tion of beef liver lies at the macromolecular
level, with the remaining one-third due to struc-
tural features of the tissue. Since the primary
mechanism for the absorption of sound in tissue
seems to occur at the macromolecular level, it
has been instructive to investigate the acoustic
properties of simpler systems of macromolecules,
such as solutions of biopolymers, with the hope
that such studies may provide details applicable
to the more complex systems.
Three measurement systems have been used at the
Bioacoustics Research Laboratory to study acoustic
absorption and velocity of solutions and suspen-
sions of biomacromolecules as functions of mole-
cular characteristics, pH, temperature, concentra-
tion, and frequency to aid in the elucidation of
the mechanisms of interaction between ultrasound
and biological tissues.
A. Absorption: High Frequency Method
(5 MHz to 200 MHz)
The high frequency system applies pulsed ultra-
sound to solutions for the purpose of measuring
absorption and velocity [24-26]. Two advantages
of pulsed ultrasound are that heating effects and
standing waves are virtually eliminated. Heating
is proportional to the mean power. Using a 10 ps
pulse at a repetition frequency of 300 pps results
in a signal which is present for only 0.3 percent
of the time. Thus the average intensity is much
less than the peak intensity, and produces a
negligible temperature rise. Since the signal is
on for such a small fraction of the time, standing
waves are not produced. The basic technique was
first described by Pellam and Gait in 1946 [27].
Two transducers, arranged coaxial ly face each
other at opposite ends of a cylindrical tank. The
receiving transducer has its position fixed while
the sending transducer moves toward or away from
it at a constant speed. Assuming the relation
governing the process is
p(x) = p(o) e"
(6)
where p(x) is the wave pressure amplitude at the
distance x from the transmitting transducer, p(o)
is the pressure amplitude at x = o (i.e., at the
face of the transmitting transducer), and a is the
absorption coefficient, the natural logarithm of
the ultrasonic pressure amplitude is proportional
to the signal path length. The proportionality
constant, i.e., the absorption coefficient, can
be found by monitoring the pressure as a function
of intertransducer distance. A block diagram of
the system is shown in figure 2. A pulsed rf
signal is amplified, passed through a variable
FREQUENCY
SYNTHESIZER
PULSE
DOUBLE-
BALANCED
MIXER
GENERATOR
WIDE BAND
AMPLIFIER
HATCHING
NETWORK
TRANSMITTING
TRANSDUCER
RECEIVING
TRANSDUCER
T
PREAMPLIFIER
SPECTRUM
ANALYZER
OPERATIONAL
AMPLIFIER
PULSE
HEIGHT
DETECTOR
CHART
RECORDER
1
ISOLATION
AMPLIFIER
DIGITAL
VOLTMETER
DIGITAL
COMPUTER
Fig. 2. Block diagram of instrumentation used for
high frequency (5 to 200 MHz) ultrasonic
absorption measurements in biological
liquids.
46
attenuator, and impedance-matched to the sending
transducer. Both transducers are air backed,
gold-on-nickel plated X-cut quartz with a funda-
mental resonant frequency of, for example, 3 MHz,
and operated at their odd harmonics. The wave
passes through the liquid, where it is attenuated,
and impinges upon the receiving transducer. Some
of the ultrasound is then reconverted into an
electric signal, while the remainder is reflected
back through the liquid. A pulse height monitor
continually detects the voltage peak of the
primary signal (which is linearly proportional
to p(x)) as the transducers are moved toward, or
away from each other, and converts it to its
natural logarithmic value. An on-line computer
samples the output of the pulse height detector
(a dc voltage proportional to the pulse height)
via a digital voltmeter, calculates the attenua-
tion coefficient, and corrects for diffraction
effects by the method of Del Grosso [28] yielding
the true absorption coefficient. Solute absorp-
tion is obtained by subtracting results of a sol-
vent measurement from a solution measurement.
Approximately one liter of sample is required for
this technique, with measurement error in absorp-
tion of about 5 percent over the approximate
frequency range 3 to 200 MHz. Greater errors oc-
cur in low absorbing liquids at low frequencies
and are mainly due to relatively large diffraction
effects.
B. Velocity: High Frequency Method
The acoustic velocity can be determined in the
high frequency system to within 0.1 percent by
measuring the period of the wave in the liquid.
This is accomplished by superimposing a reference
signal on the received signal, which results in
an interference pattern. If the acoustic path
length is changed at a constant rate, Vp, the
period of this pattern is T = (x/Vp) where X is
the acoustic wavelength. Since X - c/f, where f
is the excitation frequency, the acoustic velocity
positioning rod
is given as c = TfVp. The period T is measured
using a time interval counter to determine the
time required for the interference signal to go
through a greater number (100) of maxima.
C. Absorption: Low Frequency Method
(0.3 MHz to 20 MHz)
In the frequency range of 0.3 to 20 MHz, a
pulse technique developed by Schwan and Carsten-
sen [29] is employed in order to avoid having to
make unreasonable corrections for effects of dif-
fraction phenomena. A sending and a receiving
transducer face each other; the former is in the
reference liquid and the latter in the test liquid
(see fig. 3). An acoustic window separates the
two liquids. The distance between the trans-
ducers remains constant while the entire trans-
ducer ensemble is moved horizontally at a constant
speed from a position such that the acoustic path
is primarily in the test liquid to a position such
that it is primarily in the reference liquid. The
reference liquid is chosen to be dispersionless
and acoustically well characterized with a sound
velocity as near to that of the unknown as pos-
sible. Water is an appropriate choice for dilute
aqueous solutions. Since the path length is
constant, diffraction effects, due to the in-
equality of the velocities in the two liquids,
are corrected by a computer using the method of
Del Grosso [28]. Liquid volumes of 1 to 4 liters
are used in each chamber, depending upon frequency.
As with the high frequency system, measurement
error ranges from 2 to 10 percent. The pressure
amplitude in the receiving transducer is given by
-aw(d-
(7)
where p^ is the pressure amplitude at the sending
transducer, and the other symbols are described
by figure 3. This equation can be rewritten as
1 n p„ = In Pt
o, d +
w
(8)
transducer
acoustic window
X
transducer
Figure 3. Schematic diagram of apparatus used for ultrasonic absorption and velocity measurements
at low frequencies (0.3 MHz to 20 MHz) in biological liquids [24].
47
and can be obtained graphically from the slope
since is known.
As the acoustic window (polyethylene) reflects
a negligible, but constant amount of the acoustic
energy, it will affect the results only by chang-
ing the intercept, i.e., the slope is not affect-
ed. Signal processing to obtain the desired
absorption coefficient closely resembles that
used in the high frequency system, described
above [14,24].
nally developed by Eggers [31], which employs two
X-cut quartz disks, with a fundamental frequency
of the order of 2 MHz, separated by a lucite ring,
and forming a cylindrical cavity in which the
specimen solution is placed. One transducer
serves as a transmitter, and the other a receiver.
When the transmitting transducer is excited with a
sinusoidal voltage, standing waves in the specimen
solution result at particular frequencies that
obey the resonance equation
D. Velocity: Low Frequency Method
Velocity of the acoustic wave can be determined
with the low frequency system to 0.01 percent
[30] by superimposing the received and reference
signals in the same fashion as was done for the
high frequency system. If the velocities in the
test and reference liquids are different, moving
the transducer assembly will change the acoustic
path length. By observing the interference pat-
tern on the oscilloscope, one can position the
transducers such that a maximum occurs. If this
is position 1 in figure 3, the number n, of
acoustic wave lengths between the transducers is
given by
n=^^ + ^ (9)
If the transducers are moved a distance. Ax, such
that the interference pattern undergoes an inte-
gral number, m, of 2t\ phase shifts, the above ex-
pression becomes
d - X + Ax , x
n ± m = + -
Ax
(10)
where the positive sign applies when the velocity
in the test liquid is larger than that of the re-
ference liquid, as is usually the case. Eliminat-
ing n between these equations, expressing wave-
lengths in terms of velocity and frequency, and a
rearranging of terms yields
Pq^q
Cntan
PcC
s^s
r-tan 1
[cotanj
2 f.
(12)
where pq and c„ are the density and speed of '
sound in the quartz transducers, and Pg and c^
are the corresponding quantities in the solution.
The fundamental frequency of the transducer is
fg, and of the liquid-filled cavity, f^. At the
resonant frequencies fp, the receiving trans-
ducer registers pronounced voltage peaks, whose
frequency separation depends on the sound velo-
city of the specimen solution, and the half-
power (3 dB) bandwidth Af of which is related to
the attenuation per wavelength, aX. Eggers
[31,32] has shown that the velocity of sound Vf
in an unknown medium may be related to the velo-
city of sound in some reference medium v,, by the
expression
v_
D
1 + 2(DfZf - D^Ir)
^q^q
(13)
where If and Z„ are respectively, the acoustic
impedance of tne unknown and reference medium,
and and Dp are the respective separations in
frequency units between adjacent resonances for
the liquid and reference media, and f^ and I„
respectively, are the frequency of the quartz and
its acoustic impedance. Equation 13 may be ap-
proximated by a simpler expression (with a dif-
ference of only a few parts per thousand) as
Cv =
(11)
fAx
Note that some other procedure must be performed
to determine the correct sign.
E. Absorption and Velocity Measurements
in Small Liquid Volumes:
Resonant Cavity Method
The various methods for determining the acoustic
propagation properties of liquids discussed thus
far were devised with little consideration for the
volume of material necessary for measurement. The
minimum volume required in the high frequency
system is 500 ml, with greater volumes generally
required at lower frequencies to avoid effects due
to diffraction phenomena. The necessity for having
such large volumes available becomes a serious
problem when biological macromolecules are to be
treated in solutions (usually in concentrations of
10 percent by weight), since only a few can be
examined within the bounds of reasonable economics.
It is possible to reduce the specimen volume to
less than 50 ml using a resonance technique or^'^i-
&Cf
6f,
(14)
where SCf is the velocity difference between the
unknown and reference media, Sf^ is the difference
between corresponding resonances, and Cf is the
velocity of the reference medium at a frequency
fp. In general, velocity measurements are more
difficult to perform than absorption measurements
due to temperature drift and other instabilities
of the unit. The attenuation of the specimen
solution may be calculated from the examination
of the quality factor Q of the liquid filled
cavity, defined as the frequency fp divided by the
3 dB bandwidth, Afp, of the resonance. This quali
ty factor is a function of the mechanical clamping
and the attenuation per wavelength aX of the ultra
sonic energy and is expressed as.
Q =
Afn
oX
(15)
The measured Q, however, includes losses associat-
ed with attenuation in the solvent, as well as ^
those arising from diffraction, wall effects, im-
perfect reflections at the quartz surface, etc. ,
48
in addition to the desired excess attenuation due
to the solute. Assuming all of these energy losses
are additive [32], the measured quality factor Q
is given by
%eas ^solute ^extra
where Qsolute is the quality factor due to sound
absorption in the solute, and Qextra includes sol-
vent and other cell losses previously discussed.
The excess solute absorption can be obtained by
means of a reference measurement in the same cell
at the same frequencies with the reference liquid,
having equal or very similar sound velocity to in-
sure the same sound field pattern for both mea-
surements. The excess absorption per wavelength
in the specimen solution is then obtained from
-lexcess \ ^n /
where Af^ and Af^ are the corresponding 3 dB
bandwidtns of the n^h resonant peak in the sample
and reference liquid, respectively. The applica-
tion of an overpressure (usually about 10 psi)
causes a slight concavity of the transducers,
which reduces diffraction and boundary effects,
and thereby reduces the minimum frequency of
measurement [33]. This technique has been used
in our laboratory over the frequency range from
0.5 to 10 MHz.
5. Absorption and Velocity Measurements
of Shear Waves in Biological Specimens
A system based on a pulse superposition tech-
nique allows measurement of the shear acoustical
properties of biological materials of interest
[34]. The technique involves phase-amplitude
balance to measure the magnitude and phase angle
of the reflection coefficient of a shear wave
impinging on a quartz-sample interface, and using
the known impedance of the quartz, the complex
shear specific impedance of the sample being in-
vestigated is obtained [35]. From the shear im-
pedance and density, it is possible to calculate
the dynamic shear stiffness pi, dynamic shear
viscosity \i2' shear velocity cs, and shear absorp-
tion coefficient ag as shown in the following
relationships:
impedance and Xs is the imaginary part of the
specific shear acoustic impedance. This method
has been used to measure the shear acoustic prop-
erties of tissues and other biological specimens
[36-38]. An improved system utilizing a gated
carrier as opposed to a pulsed oscillator approach
and an improved ultrasonic unit of beveled AT
quartz is being developed that will be useful
over the frequency range from 2 to greater than
20 MHz and for temperatures from 10 to 40 °C.
These and additional improvements should provide
accuracies for the real and imaginary parts of the
specific acoustic impedance v/ithin ± 100 mech-ohm/cm^
at 10 MHz, allowing more accurate measurement of im-
pedance close to that for water.
6. Concluding Remarks
The measuring techniques for ultrasonic absorp-
tion, attenuation and velocity in biological speci-
mens discussed above have been only those utilized
at the Bioacoustics Research Laboratory. There
are, however, techniques employed elsewhere which
may provide the same information in a more effi-
cient manner. For example, the above discussed
methods provide information at discrete frequen-
cies and loci, or, for spatial averages. Spec-
trum analysis methods [39-41] have been promoted
to permit measurements over a somewhat broadened
frequency range. The measurement of the spatial
distribution of ultrasonic properties such as at-
tenuation or velocity within a specimen may be
possible using the algebraic reconstruction from
two-dimensional acoustic projections [42,43], al-
though only velocity reconstruction has been con-
sidered practical so far.
Recent studies have also considered improve-
ments regarding possible error in established
methods of ultrasonic parameter measurements.
Artifacts in acoustic attenuation due to phase
cancellation effects have been described by
Marcus and Carstensen [6], who compared a piezo-
electric receiver (phase-sensitive) with a
radiation-force receiver (phase-insensitive), and
also by Busse et al . [44], comparing the piezo-
electric receiver with an acousto-electric re-
ceiver. Both studies found that measurements
using piezoelectric (phase-preserving) receivers
yielded higher apparent attenuation values than
those obtained using phase-insensitive receivers.
Phase cancellation artifacts are thought to be
the source of this error.
Pi = M_Lis (18)
P The authors acknowledge gratefully the partial
support for portions of the activities described
Y herein by grants from the National Institutes of
V2 = (19) Health.
2 Acknowledgment
up
Rc +
pojX
References
s '^s (20) [1] Rooney, J. A., Determination of acoustic
pR power outputs in the microwatt-milliwatt
range. Ultrasound Med. Biol. 1, 13-16 (1973).
^— ^^^^ [2] O'Brien, W. D. , Jr., Ultrasonic Dosimetry,
in Ultrasound: Its Application in Medicine
where p is the density, u is the angular frequency,
Rj is the real part of the specific shear acoustic
and Biology, F. J. Fry, ed. , Chap.
(Elsevier, Amsterdam, in press]
49
[3] Kossoff, G., Balance technique for the
measurement of very low ultrasonic power
outputs, J. Acoust. Soc. Am. 38^, 880-881
(1965).
[4] Hill, C. R. , Calibration of ultrasonic beams
for biomedical applications, Phys. Med.
Biol . 15, 241-248 (1970).
[5] Pohlman, R. , Uber die Absorption des Ultra-
schalls im menschlichen Gewebe und ihre
Abhangigkeit von der Frequenz, Physik Z. 40,
159- 161 (1939).
[6] Marcus, P. W. and Carstensen, E. L., Prob-
lems with absorption measurements of in-
homogeneous solids, J. Acoust. Soc. Am. 58 ,
1334-1335 (1975).
[7] Dussik, K. T., Fritch, D. J., Kyriazidou, M. ,
and Sear, R. S., Measurements of articular
tissue with ultrasound, J. Phys. Med. 37 ,
160- 165 (1958).
[8] Bauld, T. J. and Schwan, H. P., Attenuation
and reflection of ultrasound in canine lung
tissue, J. Acoust. Soc. Am. 56, 1630-1637
(1974).
[9] Wladimiroff, J. W., Craft, I. L., and Talbert,
D. G. , In vitro measurement of sound velocity
in human fetal brain, Ul trasound Med. Biol .
I, 377-382 (1975).
[10] Buschmann, W. , Voss, M., and Kemmerling, S. ,
Acoustic properties of normal human orbit
tissues, Ophtal. Res. I, 354-364 (1970).
[11] Lange, J. N., Group velocity dispersion due
to pulse reflection from a frequency depend-
ent boundary impedance, J. Acoust. Soc. Am.
40, 998-1001 (1966).
[12] Goldman, D. E. and Richards, J. R., Measure-
ment of high frequency sound velocity in
mammalian soft tissues, J. Acoust. Soc. Am.
26^, 981-983 (1954).
[13] Hueter, T. F. , Viscoelastic Losses in Tis-
sues in the Ultrasonic Range, Wright Air
Development Center Technical Report 57-706
(1958).
[14] Fry, W. J. and Dunn, F. , Ultrasound:
Analysis and Experimental Methods in
Biological Research, in Physical Techniques
in Biological Research, W. L. Nastuk, ed..
Vol. IV, Chap. 6, pp. 261-394 (Academic
Press, New York, 1962).
[15] Dunn, F. and Fry, W. J., Ultrasonic absorp-
tion and reflection by lung tissue, Phys.
Med. Biol. 5^, 401-410 (1961).
[16] Fry, W. J. and Fry, R. B. , Determination
of absolute sound levels and acoustic ab-
sorption coefficients by thermocouple
probes--theory , J. Acoust. Soc. Am. 26,
294-310 (1954). ^
[17] Fry, W. J. and Fry, R. B., Determination
of absolute sound levels and acoustic ab-
sorption coefficients by thermocouple
probes--experiment, J. Acoust. Soc. Am. 26,
311-317 (1954).
[18] Dunn, F., Temperature and amplitude depend-
ence of acoustic absorption in tissue, J_.
Acoust. Soc. Am. 34, 1545-1547 (1962).
[19] Dunn, F. and Brady, J. K., Temperature and
Frequency Dependence of Ultrasonic Absorp-
tion in Tissue, in Proceedings 8th Inter-
national Congress on Acoustics, Vol. 1,
p. 366c (Goldcrest Press, Trowbridge ,
Wilts., 1974).
[20] Brady, J. K., Goss, S. A., Johnston, R. L.,
O'Brien, W. D., Jr., and Dunn, F., Ultra-
sonic propagation properties of mammalian
testes, J. Acoust. Soc. Am. 60_, 1407-1409
(1976).
[21] Dunn, F. and Breyer, J. E., Generation and
detection of ultra-high-frequency sound in
liquids, J. Acoust. Soc. Am. 34, 775-778
(1962).
[22] Carstensen, E. L., Li, K. , and Schwan, H. P.,
Determination of the acoustic properties of
blood and its components, J. Acoust. Soc.
Am. 25, 286-289 (1953).
[23] Pauly, H. and Schwan, H. P., Mechanism of
absorption of ultrasound in liver tissue,
J. Acoust. Soc. Am. 50, 692-699 (1971).
[24] O'Brien, W. D. , Jr., The Absorption of
Ultrasound in Aqueous Solutions of Biologi-
cal Polymers, Ph.D. Thesis, University of
Illinois, Urbana, Illinois (1970).
[25] Goss, S. A., The Concentration Dependence
of the Ultrasonic Absorption in Aqueous
Solutions of Bovine Serum Albumin, M.S.
Thesis, University of Illinois, Urbana,
Illinois (1974).
[26] Kessler, L. W., Hawley, S. A., and Dunn, F. ,
Semi-automatic determination of ultrasonic
velocity and absorption in liquids,
Acustica 24, 105-107 (1971).
[27] Pellam, J. R. and Gait, J. K., Ultrasonic
propagation in liquids: I. Application of
pulse technique to velocity and absorption
measurements at 15 megacycles, J. Chem.
Phys . 14' 608-614 (1946).
[28] Del Grosso, V.A., Systematic Errors in Ultra-
sonic Propagation Parameter Measurements
Part I--Effect of Free-Field Diffraction,
U.S. Naval Research Laboratory Report 6026
(1964).
[29] Schwan, H. D. and Carstensen, E. L., Ultra-
sonics aids diathermy experiments. Elec-
tronics 25, 216-220 (1952).
[30] Carstensen, E. L., Measurement of dispersion
of velocity of sound in liquids, J. Acoust.
Soc. Am. 26, 858-861 (1954).
[31] Eggers, F., Eine Resonatormethode zur Be-
50
stimmung von Schal 1 --Geschwindigkei t und
Dampfung an gen'ngen Fliissigkeitsmengen,
Acustica 19. 323-329 (1968).
[32] Eggers, F. and Funck, Th., Ultrasonic mea-
surements with milliliter liquid samples in
the 0.5-100 MHz range, Rev. Sci. Instrum.
44, 969-977 (1973).
[33] Eggers, F., Funck, Th., and Richman, K. H.,
High Q ultrasonic liquid resonators with
concave transducers. Rev. Sci. Instrum. 47,
361-367 (1976).
[34] Frizzell, L. A., Carstensen, E. L., and
Dyro, J. F. , Shear properties of mammalian
tissues at low megahertz frequencies, J^.
Acoust. Soc. Am. 60, 1409-1411 (1976).
[35] Moore, R. S. and McSkimin, H. J., Dynamic
Shear Properties of Solvents and Polystyrene
Solutions from 20 to 300 MHz, in Physical
Acoustics, W. P. Mason and R. N. Thurston,
eds.. Vol. VI, Chap. 4, pp. 167-242 (Academic
Press, Inc., New York, 1970).
[36] Dyro, J. F., Ultrasonic Study of Material
Related to Atherosclerotic Plaque--Dynamic
Viscoelastic Properties of Cholestoric
Esters, Ph.D. Thesis, University of Penn-
sylvania, Philadelphia, Pennsylvania (1972).
[37] Dyro, J. F. and Edmonds, P. D., Ultrasonic
absorption and dispersion in cholesteryl
esters, Mol. Cryst. Liq. Cryst. 25, 175-193
(1974).
[38] Frizzell, L. A., Ultrasonic Heating of
Tissues, Ph.D. Thesis, University of Rochester,
New York (1975).
[39]- Chivers, R. C. and Hill, C. R., Ultrasonic
attenuation in human tissue. Ultrasound Med.
Biol. 2, 25-29 (1975).
[40] Lizzi, F., Katz, L., St. Louis, L., and
Coleman, D. J., Applications of Spectral
Analysis in Medical Ultrasonography, Ultra-
sonics 14, 77-80 (1976).
[41] Le Croissette, D. H. and Heyser, R. C,
Attenuation and Velocity Measurements in
Tissue Using Time Delay Spectrometry, in
Ultrasonic Tissue Characterization, M.
Linzer, ed.. National Bureau of Standards
Spec. Publ. 453, pp. 81-95 (U.S. Government
Printing Office, Washington, D.C., 1976).
[42] Greenleaf, J. F., Johnson, S. A., Lee, S. L.
Herman, G. T., and Wood, E. H., Algebraic
Reconstruction of Spatial Distributions of
Acoustic Absorption Within Tissue From Their
Two-Dimensional Acoustic Projections, in
Acoustical Holography, P. S. Green, ed..
Vol . 5, pp. 591-603 (Plenum Press, New York,
1974).
[43] Greenleaf, J. P., Johnson, S. A., Samayoa,
W. F., and Duck, F. A., Algebraic Reconstruc
tion of Spatial Distributions of Acoustic
Velocities in Tissue From Their Time of
Flight Profiles, in Acoustical Holography,
N. Booth, ed.. Vol. 6, pp. 71-90 (Plenum
Press, New York, 1975).
[44] Busse, L. J., Miller, J. G., Yuhas, D. E.,
Mimbs, J. W., Weiss, A. N., and Sobel, B. E.
Phase Cancellation Effects: A Source of At-
tenuation Artifact Eliminated by a CdS
Acoustoelectric Receiver, in Ultrasound in
Medicine, D. White, Ed., Vol. 3, pp. 1519-
1535 (Plenum Press, New York, 1977).
51
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed., National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
A DEVICE FOR MEASURING ULTRASONIC PROPAGATION VELOCITY IN TISSUE
Bruce D. Sollish
Harry De Jur Diagnostic Instrumentation Laboratory
Department of Electronics
Weizmann Institute of Science
Rehovot, Israel
This paper discusses a device capable of measuring propagation velocity in tissue,
both excised and in vivo. The device produces a 4-digit decimal readout of propagation
velocity in the specimen relative to that in water. As presently constituted the de-
vice is capable of 1 velocity measurement per second, with one additional second re-
quired for printout of the result.
The theory of the device is described, and experimental results are presented for
solids and soft tissues.
Key words: Propagation velocity; reflection technique; solids; tissue; ultrasound.
1. Introduction
Ultrasonic propagation velocity has been mea-
sured in a variety of soft tissues. In a review
article, Wells [iy has compiled some of the more
significant velocity data published in the litera-
ture. It would appear that accurate measurement
of velocity could help in characterizing soft
tissues. Kossof f et al . [2], for example, have
measured ultrasonic velocity in the human female
breast and have concluded that the results as to
characteristics of tissue type correlate reason-
ably well with the findings of mammography.
One problem in velocity measurement is the
need for determining thickness of the tissue
along the path of propagation. Thickness can be
measured directly or by the method of equivalent
water path given by Kossoff [2,3] for transmis-
sion and reflection velocity measurement tech-
niques. For each new propagation path through
the tissue, another thickness measurement must
be made. The process of measuring velocity in
a variety of paths through the tissue could
therefore be quite time consuming.
In order to eliminate the need to determine
tissue thickness along the propagation path, a
new pulse-echo method for measuring ultrasound
velocity has been developed. This technique
utilizes a reflecting surface at a fixed dis-
tance from the transducer. The overall propaga-
tion path consists of a suitable medium, such
as water, and the tissue sample.
Since only the overall distance between the
transducer and reflector need be constant, the
thickness of the tissue need not be measured.
The technique can therefore be readily applied
in velocity measurements in different samples or
through different paths in the same sample.
^Figures in brackets indicate literature
references at the end of this paper.
without repeated thickness calculations. Indeed,
the velocity data can be used to compute thick-
ness in the specimen along the propagation path,
which can be compared with the thickness as
measured directly, for checking the accuracy of
the measurement.
A prototype of a device incorporating this
reflection technique has been built, for mea-
surements of propagation velocity in solids and
in tissue. The following sections describe in
more detail the reflection technique, the proto-
type device implementing the technique, and re-
sults of velocity measurements in solids and in
tissue.
2. Description of the Technique
The method used for measuring propagation
velocity is illustrated in figure 1. A flat
reference reflecting surface is positioned at a
convenient fixed distance from the ultrasonic
transducer. Both are immersed in a suitable
propagating medium such as distilled water. An
A-mode pulse-echo device displays an echo due to
the reflecting surface. This echo is a reference
echo for the remainder of the velocity measure-
ment.
The tissue specimen is next interposed between
the transducer and the reflector as shown in the
figure. The A-scope is adjusted to display the
echoes corresponding to the following interfaces:
the anterior boundary of the specimen, the pos-
terior boundary of the specimen, and the re-
flector surface as returned through the specimen.
The reflector echo is shifted in time from its
reference position if the velocity in the speci-
men is different from that in the medium.
As shown in the figure, there are in general
two configurations for velocity measurement. In
the first configuration, the specimen is posi-
tioned somewhere in between the transducer and
the reflector. In this case four echoes of in-
53
transducer
water
reference reflector
initial pulse
transducer
reference echo
initio! pulse
transducer
reference reflector
onginol reference
echo
displaced reference
echo
target reference reflector
n-t, J
Tllb
reference echo
initial pulse
anterior posterior / displaced
target target / reference
echo echo / echo
Fig. 1. Reference reflector technique for velocity
measurements .
a) initial calibration
b) target interposed between transducer
and reference reflector
c) target in contact with reference
ref 1 ector .
terest are produced. In the second configuration,
the specimen is positioned so that its posterior
boundary is coincident with the reflector. In
this case, there are only three echoes to con-
sider. The second configuration is readily
adapted for use in velocity measurements made via
a water bag, in which case the specimen and the
reflector lie outside the water bag.
For either configuration, velocity in the
sample is calculated as follows. Denote the dis-
tance between the transducer and reflector by L,
the thickness of the specimen by i, the velocity
in the medium by Vq, and the velocity in the
specimen by V. Let the time between the anterior
and posterior echoes from the specimen be T^, and
let the time displacement of the reflector echo
due to interposition of the specimen be T2. Then
the following equations apply for Tj and T2:
Ti = 2e/V
- 2L
Vo
T, = ^
[20:
L Vn
21"
V
= 2? i
I
Vo
(1)
(2)
From these equations, the ratio of propagation
velocity in the specimen to that in the medium is
V/Vo = 1 + T2/T1
(3)
We note that To is positive if the reflector echo
is displaced toward the transducer when the speci-
men is in position, and negative if the reflector
is displaced away from the transducer.
The thickness of the specimen along the path
of propagation can be calculated from the above
equations, as
where
5./f.o = V/Vr
VoTi
2
(4)
. (5)
We note that Sq is the thickness of the specimen
measured by the A-scope calibrated for velocity
in the medium.
The significance of eqs. (1) through (5) is
that the ratio of T2 to T^ is independent of the
thickness i of the specimen. Accurate measure-
ment of times Tj and T2 will give an accurate
measurement of velocity in the specimen relative
to velocity in the medium. Even if the specimen
is repositioned or if a different specimen is
examined, Tj and T2 will both change accordingly,
as long as the transducer to reflector distance
remains unchanged. Thus, the technique enables
rapid velocity measurements in a variety of tis-
sues, as long as the reference echo is returned
through the specimen.
3. Description of the Device
A block diagram of a device utilizing the re-
ference reflector technique for measuring ultra-
sound velocity is shown in figure 2. An A-scope
and an oscilloscope are required. In addition,
a printer may be added for hard copy output of
velocity data.
sync out \
a-scope
tinning
circuitry
program
circuitry
data
acquisition
circuitry
calculator
circuitry
digital
display
transducer
auxiliary
display
oscilloscope
Fig. 2. Block diagram of velocity measurement
device and associated equipment.
Operation of the system is as follows. Before
the specimen is immersed in water, the A-scope is
adjusted to display the echo due to the reference
reflector. The A-mode ("video") signal is fed to
the data acquisition circuit, passing through a
54
time-gated threshold detector and a pulse shaper.
The processed echo from the reflector is dis-
played on the oscilloscope, together with a pulse
whose position on the display is variable. The
user presets the time delay of the variable pulse,
so that the pulse overlaps the reference echo, as
seen on the oscilloscope and confirmed by a front-
panel LED indicator. This procedure, which
stores in the data acquisition circuitry the prop-
agation time from the transducer to the reflector,
is performed only once before the specimen is in-
serted. At any time, calibration may be checked
by removing the specimen, and noting if the over-
lap condition is resumed.
Following storage of the reference echo, the
specimen is interposed between the transducer and
the reflector. The data acquisition circuit, by
adjustment of the time-gated threshold detector,
eliminates internal echoes from within the sample.
The present velocity device implements the con-
figuration of figure Ic. Therefore only two
echoes remain after processing: the anterior
echo of the specimen and the echo at the inter-
face between the specimen and reflector.
The data acquisition circuit now contains
three pulses: the two echoes processed from the
tissue specimen and the preset pulse supplied by
the user as described earlier. Two gates are de-
rived from these pulses. The first gate is
opened by the anterior pulse and closed by the
specimen-reflector boundary echo. The second
gate is opened by the specimen-reflector boundary
echo and closed by the preset pulse or opened by
the preset pulse and closed by the specimen-
boundary echo, depending on which signal arrives
first. In the latter case, a flag is sent out
to the calculator circuit to perform a subtrac-
tion instead of an addition, as discussed later.
Each gate enables one of two 4-digit BCD
counters clocked at a rate of 15 MHz. By the end
of the second gate, two data words have been
loaded into the counters. These two 4-digit
words, corresponding to Tj and T2 in eq. (3),
provide all the information required for calcu-
lating velocity in the specimen (relative to
water). The maximum count of 9999 corresponds
to a round-trip travel time through 500 mm of
water.
A 10-step hard-wired program sequentially con-
trols the operation of a 4-function calculator
chip. These steps include clearing the calcu-
later, entering the values Ti and T?, performing
the arithmetic operations of eq. (3), and dis-
playing the result to 4 significant digits via
a front-panel LED display. In the event of a
flag generated by the data acquisition circuitry,
the program treats T2/T1 as a negative number.
If desired, the results of each measurement are
printed by an IBM output writer, 10 measurements
to a line.
The timing> circuitry maintains synchronization
among all elements of the velocity measuring sys-
tem. When the user presses the start button
after checking on the oscilloscope that the cor-
rect echoes are present, the timing circuit in-
sures that only a single set of data words enters
the calculator until calculation and printout of
the results are complete. Using a free-running
start signal, a new velocity value is calculated
and printed every two seconds.
4. Experimental Results
A. Measurements of Velocity in Solids
Before conducting measurements of ultrasound
velocity in tissues, it was necessary to check the
accuracy of the reference reflector velocity tech-
nique and its implementation in the device de-
scribed in the previous section. For this pur-
pose three objects were chosen: a 4-mm thick
slab of polyvinylchloride (PVC), a 6-mm thick
slab of perspex, and a 10-mm thick combination of
the two materials. The ultrasonic frequency was
5 MHz. Fifty velocity measurements were made for
each object.
Results of measurements of velocity in the
three test objects are given in table 1. Absolute
values of velocity are found by multiplying the
respective relative values by the ultrasound
velocity in water, 1498 m/s. The standard devia-
tions vary from about 0.3 to 0.5 percent of the
measured values.
Table 1. Velocity measurements of selected solids.
Polyvinylchloride (PVC)
Relative velocity data {- 10"')
486
1486
1479
1472
1486
1479
1479
1486
1486
1486
486
1486
1500
1500
1486
1486
1479
1479
1486
1479
486
1486
1500
1486
1486
1500
1486
1486
1486
1500
486
1500
1486
1486
1486
1486
1472
1479
1500
1479
500
1486
1486
1486
1486
1472
1472
1486
1486
1486
Average relative velocity: 1.4860
Average absolute velocity: 2226.0 m/s
Standard deviation; 11.2 m/s
Perspex
Relative velocity data
10-
1822
1811
1811
1811
1811
1811
1811
1822
1822
1811
1811
1820
1811
1811
1822
1811
1811
1820
1811
1811
1811
1820
1820
1800
1811
1822
1811
1811
1808
1800
1822
1811
1811
1800
1820
1811
1811
1811
1822
1820
1811
1811
1820
1820
1811
1811
1831
1811
1820
1811
Average relative velocity: 1.8138
Average absolute velocity: 2717.1 m/s
Standard deviation: 9.4 m/s
PVC-perspex combination
Relative velocity data (« 10" 3)
1670 1664 1664 1668 1674 1668 1674 1668 1668 1664
1664 1658 1664 1668 1664 1664 1670 1670 1674 1664
1658 1664 1664 1664 1664 1664 1664 1674 1658 1664
1658 1664 1674 1674 1670 1674 1668 1664 1664 1664
1668 1668 1668 1664 1674 1658 1674 1658 1668 1668
Average relative velocity: 1.6664
Average absolute velocity: 2496.3 m/s
Standard deviation; 7.2 m/s
To check the accuracy of the results obtained
for each, object, the thickness given by eqs. (4)
and (5) was compared to the thickness measured by
a caliper. Propagation times were respectively
3.60 ps, 4.42 us, and 8.02 us in the PVC, perspex,
and PVC-perspex combination. The corresponding
thicknesses calculated from the data and eqs. (4)
and (5) are 4.003 mm, 6.005 mm, and 10.01 mm,
respectively, which are within 0.1 percent of the
measured values.
55
B. Measurements of Velocity in Soft Tissues
In order to demonstrate measurement of velocity
in soft tissue, a sample of fat in vitro and two
human forearms in vivo were examined. As previ-
ously, the insonifying frequency was 5 MHz and 50
individual velocity measurements were taken for
each target. Results of the measurements are
given in table 2.
Table 2. Velocity measurements of soluble soft tissues.
Fat (bovine)
Relative velocity data (> 10"')
0987 0987 0987 0987 0987 0987 0987 0987 0991 0991
0991 0991 0991 0987 0987 0991 0987 0987 0987 0991
0987 0987 0987 0991 0987 0991 0991 0991 0987 0987
0991 0987 0991 0987 0991 0991 0987 0991 0983 0991
0991 0987 0987 0987 0991 0987 0987 0987 0987 0987
Average relative velocity: 0.9884
Average absolute velocity; 1480.7 m/s
Standard deviation: 3.1 m/s
Forearm (subject A)
Relative velocity data (» 10"^)
1057 1058 1058 1058 1057 1057 1058 1057 1057 1056
1056 1057 1057 1057 1057 1057 1058 1057 1058 1058
1057 1057 1057 1057 1056 1057 1057 1057 1057 1057
1056 1057 1056 1056 1057 1057 1057 1057 1057 1057
1057 1058 1057 1059 1057 1057 1057 1057 1056 1056
Average relative velocity: 1.0570
Average absolute velocity: 1583.4 m/s
Standard deviation: 1.0 m/s
Forearm (subject B)
Relative velocity data (■ 10"')
1056 1056 1055 1055 1056 1056 1055 1056 1055 1055
1056 1055 1055 1056 1056 1055 1056 1056 1056 1055
1055 1056 1055 1055 1056 1055 1054 1053 1054 1053
1053 1053 1053 1054 1054 1055 1054 1054 1054 1053
1054 1054 1054 1053 1054 1053 1053 1054 1053 1053
Average relative velocity: 1.0546
Average absolute velocity: 1579.8 m/s
Standard deviation: 1.7 m/s
that velocity measurements accurate to within
0.2 percent can be achieved with the device.
An improved model is now under construction.
It will incorporate microprocessor-based circuit-
ry for increasing the number of measurements at
a single location to 500 per second. This will
increase the precision of the measured velocity
as well as eliminate problems of motion during
measurement. The new device should enable scan-
ning a structure such as the breast with the
object of studying local variations in propaga-
tion velocity.
Acknowledgments
The author wishes to express his gratitude to
the Harry de Jur Foundation and the Rose Teitel-
baum Cancer Research Foundation for their par-
tial support of this research. The author would
also like to thank Professor E.H. Frei, and Y.
Dreier, I. Gonen, E. Grinwald, A. Kuprak, J.
Leibovitz, and M. Moshitzky of the Weizmann
Institute for their roles in the project.
References
[1] Wells, P. N. T., Absorption and dispersion
of ultrasound in biological tissue. Ultra-
sound in Med, and Biol. U 369-376 (1975).
[2] Kossoff, G., Fry, E. K. , and Jellins, J.,
Average velocity of ultrasound in the human
female breast, J. Acoust. Soc. Am. 53 ,
1730-1736 (1973T:
[3] Kossoff, G., Reflection Techniques for Mea-
surement of attenuation and velocity, in
Ultrasonic Tissue Characterization, M. Linzer,
ed., National Bureau of Standards Spec. Publ.
453, pp. 135-139 (U.S. Government Printing
Office, Washington, D.C., 1976).
The specimen of bovine fat was refrigerated
for several days and then warmed to room tempera-
ture before insoni fi cation . The velocity of
1481 m/s is approximately 1 percent higher than
the velocity for cow fat given by Goldman and
Hueter [4], but well within the limits for mam-
malian fat in general. The standard deviation
in the present measurement is about 0.2 percent.
The next measurements were made in the fore-
arms of two male subjects, the same age (27).
Care was taken to investigate the same portion
of forearm of each subject. The average veloci-
ties in the two subjects, 1583 m/s and 1580 m/s,
respectively, lie within the range given by
Goldman and Hueter for human limbs. The standard
deviations are of the order of 0.1 percent of the
measured velocities. More data would be required
to determine if the difference in velocity be-
tween the two subjects is statistically signifi-
cant.
[4] Goldman, D. E., and Hueter, T. F., Tabular
data on the velocity and absorption of high
frequency sound in mammalian tissues, J^.
Acoust. Soc. Am. 28, 35-78 (1956).
5. Conclusion
A device for measuring ultrasonic propagation
velocity in tissue using a reference reflector
has been described. On the basis of experimental
results in solids and soft tissue, it appears
56
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed. , National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
MEASUREMENT OF THE TEMPERATURE DEPENDENCE OF THE VELOCITY
OF ULTRASOUND IN SOFT TISSUES
T. Bowen, W. G. Connor, R. L. Nasoni , A. E. Pifer, and R. R. Sholes
Departments of Physics and Radiology
University of Arizona
Tucson, Arizona 85724, U.S.A.
The velocity of 5 MHz ultrasound is being measured in tissue samples in order
to evaluate the feasibility of non-invasive monitoring of temperature distributions
produced during hyperthermia treatments. By employing a pul sed-ul trasound tech-
nique in which the time of the first zero-crossing of the received signal is re-
corded, the velocity measurements are insensitive to reflections and to changes of
attenuation. Results in the 35 to 45 °C range for fresh canine tissues, such as
kidney, liver, and muscle, indicate that the rate of velocity change with tempera-
ture is correlated with the magnitude of the ultrasonic velocity, but the relation-
ship appears to be altogether different for tissue fat. The results give encourage-
ment to carry out the more extensive measurements, particularly in vivo, which would
be needed to determine feasibility of an ultrasonic temperature monitoring system.
Key words: Non-invasive temperature monitoring; soft tissue; temperature dependence;
ultrasound velocity; zero crossing detection.
1. Introduction
There has been increasing interest recently
in hyperthermia as a promising modality for con-
trol of some types of malignancies [1-3]^. In
such treatments the tissue temperature must be
raised from the usual body temperature of 37 °C
to approximately 43 °C for a period of one-half
to one hour. If the temperature is accidentally
a few degrees higher, all cells will be killed;
if a few degrees lower, the effect of the treat-
ment will be negligible. A non-invasive tem-
perature monitoring system is needed for this
application. In addition, such a system might
find other diagnostic applications where thermal
anomalies are caused by changes in the circula-
tory system or in metabolic processes.
An ultrasonic temperature monitoring system
might be devised for soft tissue if some ultra-
sonic characteristic varies in a known way with
temperature. The most promising parameters ap-
pear to be ultrasonic velocity and attenuation.
Earlier work on transmission-type tomographic
image reconstruction systems [4] indicated that
velocity data gives much more reproducible re-
sults. However, the task of utilizing velocity
changes to measure temperature is not an easy
one. At 37 °C the velocity of sound in water
changes 1.8 m/s per °C or about 0.12 percent
per °C. It is clear when one examines results
for the velocity of ultrasound in any particu-
iFigures in brackets indicate literature
references at the end of this paper.
lar tissue type that the variation from one
sample to another is much larger than the
variation over a 10 °C temperature range; con-
sequently, one cannot hope to measure tem-
perature ultrasonically. However, it still
might be possible to measure temperature
'"hanges ultrasonically, making use of the fact
it the initial reference temperature, 37 °C,
is known to be quite uniform and corrections
for regional differences are also known.
Suppose it is desired to monitor a tempera-
ture change from body temperature (37 °C) to
a treatment temperature in the neighborhood
of 43 °C within ± 10 percent for each tissue
type in the region of interest. It is the
purpose of this work to discuss the first
experimental step toward answering whether or
not this property of soft tissue can be pre-
dicted with such accuracy.
2. Experimental Arrangement
The experimental arrangement for the in vitro
measurement of ultrasonic velocity in 1 to 10 cm
thick tissue samples as a function of tempera-
ture is shown schematically in figure 1. The
pulse generator provides a pulse which shock
excites the 12.7 mm diameter 5 MHz ultrasound
transmitter immersed in the water temperature
bath. This pulse also starts the digital timer.
The oscilloscope serves only to visually moni-
tor the system; its trace is initiated by a
trigger pulse from the pulse generator whose
timing can be delayed or advanced relative to
the main pulse so as to observe any part of
57
Digital VOM
(Keithley 168)
Plastic
Sample Bag
Thermistor
a
Acoustic
Transmitter
r:
Acoustic
Receiver
' Heater/
Circulator
(Haake E52)
Thermometer
Tissue Sample
Temperature Bath Tank
Pulse Generator
(HP 2I4A)
Pulse Trigger
Out Out
Oscilloscope
(Tektronix R(vl45A)
Ext Vertical
Trigger Input
(a) Large Signal
Fig. 1. Block diagram of the experimental
arrangement for in vitro measure-
ments of ultrasound velocity as a
function of temperature.
the signal at the receiver. The output of the
12.7 mm diameter ultrasound receiver, which is
also immersed in the temperature bath, is con-
nected both to the vertical oscilloscope input
for observation of its amplitude and to a zero-
crossing discriminator whose output stops the
digital timer. The digital timer measures to
the nearest 0.01 ms the transit time plus a
fixed time related to the delays in the system
not connected with the sound propagation. The
fixed delays are found by an extrapolation to
zero separation of a series of measurements
of transit time in water as a function of
separation within the errors of measurement.
This agrees with the results of others who have
found that, although the pressure distribution
in the near field is very complicated, the
average pressure sensed by a detector whose
dimensions are large compared to a wavelength be-
haves like a plane wave [5]. Adjustment of the
width of the shock excitation pulse from the
pulse generator can provide a signal at the re-
ceiver in which only one or two cycles have large
ampl i tude .
The zero-crossing discriminator is an essen-
tial component to define a time reference in a
received signal burst. Its operation is illus-
trated in figure 2. By initiating discriminator
action at the occurrence of the first zero
crossing, the timing becomes independent of re-
ceived signal amplitude. This is important when
studying the temperature dependence of sound
velocity as the attenuation (hence, the received
receiver
signal
time
first zero crossing
(b) Small Signal
recei ver
signal
time
first zero crossing
Fig. 2. The first zero-crossing is shown
for a large signal (a) and a small
signal (b) to illustrate that its
position in time remains unaffeced
by changes of amplitude.
signal amplitude) also varies with temperature.
A pulsed ultrasound system with the timing ref-
erenced to the first zero crossing not only is
very stable in operation [6], but also is com-
paratively immune to the effects caused by
multiple reflections, as these only disturb a
later part of the received signal.
The frequency dependence of ultrasonic veloc-
ity and attenuation in tissue affects the shape
of the received signal, which could introduce
timing shifts. The change of velocity with fre-
quency, or dispersion, causes the pressure pulse
to propagate with a group velocity differing
slightly from the phase velocity of a continu-
ous wave. Tissue dispersion is typically 0.1
to 1.0 m/s per MHz [7], which at 5 MHz shifts
the group velocity 0.5 to 5 m/s above the phase
velocity. For sample thicknesses on the order
of 2 cm the first zero crossing is still propa-
gating with a velocity nearly equal to the
group velocity, and only the leading edge of the
first one-half cycle is distorted because the
pulse is moving faster than its constituent fre-
quency components [8]. A small velocity shift
of this magnitude would have little effect upon
the observed rate of change with temperature.
The increasing attenuation at higher fre-
quencies in tissue changes the signal waveform
with changing depth. However, the shape of the
received waveform was not observed to change
significantly over the 35 to 45 °C range of
temperatures at fixed depth.
For path lengths on the order of 2 cm the
transit time jitter is less than 0.001 \is . The
0.01 \iS least count of the digital timer pres-
ently limits the timing precision, but this can
be greatly improved with the same time digi-
tization by adding an event counter and digital
logic so that the transit time is automatically
accumulated for, say, 1000 transmitted pulses.
However, this improvement will come about only
if the time reference oscillator is not synchro-
nized with the start of the transit time meas-
urement.
58
For samples more than a few centimeters
thick, which may often be encountered for i n
vivo measurements , the optimum ultrasound fre-
quency may be somewhat less than the 5 MHz ini-
tially employed. If V(t) is the received
transducer voltage as a function of time near
the first zero crossing and if the voltage er-
ror in detecting the true zero point is given,
it is desired to maximize dV/dt at the zero
crossing. This would be proportional in the
near field to f exp (-gaf), where a is the
spacing between transducers and gf is the
ultrasonic amplitude attenuation coefficient
for the tissue being examined. This function
is maximum when gaf = 1, or f = (ga)- . For
example, 6 ^ 0.01 ps-mm"! is typical for many
soft tissues, so if a = 100 mm, f = 1 MHz.
This result can be easily generalized for cases
where the voltage error or noise is frequency
dependent.
In a typical data-taking run, the tempera-
ture of the bath is increased at a rate of 0.4
°C/miri. The rate must be reasonably high so
that changes do not take place in the tissue
which are a function only of time. However,
the higher the rate of change of temperature,
the greater the lag of temperature at the
center of the sample. In order to correct the
data for this temperature lag, a thermistor is
inserted at the center of the sample to monitor
the temperature difference with respect to the
bath. It can be shown from the solution of the
one-dimensional heat-diffusion equation when
the temperature at the surfaces is a linear
function of time that the average temperature
along the ultrasound path at time t, Q(t), is
approximately given by:
^(t) =i«bath ^ Wmistor '
"^^'^ %ath thermistor ^^^^^^^ ^he
temperatures at time t of the ,bath (and sample
surfaces) and center thermistor, respectively.
TEMPERATURE (°C)
Fig. 3. The speed of ultrasound as a function of
temperature for water. The points repre-
sent data obtained in this work. The
solid curve represents the results of J.
R. Lovett, J. Acoust. Soc. Amer. 45,
1051 (1969).
For samples more than a few centimeters thick
and for in vivo measurement, other methods of
heating the sample may be necessary to avoid
large temperature inhomogeneities. This may
often necessitate synchronized alternate puls-
ing of the heat source and the ultrasound
transmission to avoid spurious electrical
pickup.
Data obtained with the system outlined above
for water is illustrated in figure 3. Note the
excellent agreement when the rate of variation
with temperature is compared with results ob-
tained by others [ 9] .
3. Results
Freshly excised samples of canine skeletal
muscle, liver, kidney, spleen, brain, and fat
were measured within one hour after sacrifice
of the animal. Each run began at approximate-
ly 35 °C. When approximately 45 °C was reached
the temperature was held constant for at least
30 minutes to verify that the observed changes
were attributable to the changes of tempera-
ture and not to tissue degeneration. Some data
were taken where the temperature was held con-
stant at other values between 35 and 45 °C; no
runs gave evidence of ultrasound velocity
changes due to tissue degeneration for samples
within the first hour after sacrifice. At
later times (> 60 minutes after sacrifice),
velocity changes (occasionally dramatic de-
creases) at constant temperature were ob-
served with muscle samples. Each data run was
least-square fitted to a polynomial in tem-
perature containing linear and quadratic terms.
Table 1. Ultrasound velocity at 37 °C and rate
of change of ultrasound velocity with
temperature, dv/de, at 37, 40, 43 °C
for fresh canine tissue samples, water,
and corn oi 1 .
Sound
velocity (dv/do) [ (m/s )/°C]
Tissue at 37 °C
type
(m/s
37
°C
40 °C
4;
°
Skeletal muscle
1589
1
1 .
23
1
03
0.
82
Skeletal muscle
1603
3
1 .
13
0
92
0.
71
Skeletal muscle
1588
8
1 .
16
0
95
0.
78
Skeletal muscle
1591
6
1 .
08
0
87
0
65
Liver
1591
7
0.
93
0
78
0
62
Liver
1594
8
1
13
0
96
0
80
Liver
1604
0
0
99
0
72
0
46
Kidney
1570
2
1
35
1
16
0
98
Kidney
1566
5
1
29
1
11
0
93
Kidney
1571
1
1
29
1
11
0
94
Bra i n
1563
2
0
67
0
62
0
26
(white matter)
Spl een
1601
3
1
31
1
07
0
84
Water
1522
5
1
84
1
66
1
48
Stomach fat
1411
9
-2
89
-2
85
-2
86
(fresh)
91
Stomach fat
1412
9
-3
43
-2
86
-2
.(refrig. 5 h)
Corn oil
1420
0
-2
75
-2
58
59
1600
1550
skeletal
•muscl e
'1 i ver
• ••kidney
• brain
•water
1400
1350
stomach
fat
37 39 41 43
TEMPERATURE (°C)
45
Fig.
Typical data showing the dependence
of ultrasound velocity on tempera-
ture for fresh canine skeletal muscle,
liver, kidney, brain (white matter),
and stomach fat, as well as for water.
The results are listed in table 1 along with
the results for water and corn oil for compari-
son. Typical curves of ultrasound velocity
versus temperature are shown in figure 4. In
every case, the variation of ultrasound veloci-
ty with temperature, dv/de', is less than the
corresponding value for water, actually re-
versing sign for fat. For tissues other than
fat at a specified temperature, a negative
correlation between the magnitude of the ultra-
sound velocity, v, and dv/de is apparent. This
is shown in figure 5, where dv/de at 37 °C is
plotted against v at 37 °C for each sample (ex-
cept brain and fat) and for water. Except for
one data point for liver and the single point
for spleen in figure 5, there appears to be a
close connection between dv/de and v for water
and soft tissues having high water content.
4. Conclusions
The results of this study are encouraging in
two respects: (1) Changes of ultrasound veloci-
ty in soft tissues with temperature can be simp-
ly, yet accurately, measured with readily avail-
able digital electronic circuitry. (2) The rate
of change of ultrasound velocity in soft tissues
with high water content in this study fell with-
in ± 10 percent of a prediction based only upon
1.9
1 .8
1.7
o
3. 1.6
1 1.5
1.4-
3 1.3
1.2
1.1
1.0
0.9
1520
Fig. 5.
□ water
A kidney
o liver
• muscle
X spleen
A
A A
O
J I
1540
1560 1580
v (m/s)
1600
The rate of change of ultrasound
velocity with temperature, dv/de,
at 37 °C is plotted as a function
of the corresponding ultrasound
velocity, v, at 37 °C for each
fresh canine tissue sample and for
water.
the magnitude of the velocity itself for 9 out of
12 samples. However, it is already apparent that
the relative fat content of various tissues may
be an important parameter. Before any conclu-
sions can be reached concerning the feasibility
of non-invasive ultrasonic temperature monitor-
ing, further measurements, especially in vivo,
must more precisely ascertain the predictability
of ultrasonic velocity temperature dependence in
soft tissues.
References
[1] Gerner, E. W., Connor, W. G., Boone, M. L.
M., Doss, J. D. , Mayer, E. G., and Miller,
R. C, Radiology 116, 433 (1975).
[2] Miller, R. C, Connor, W. G., Heusinkveld,
R. S., and Boone, M. L. M., Prospects for
Hyperthermia in Human Cancer Therapy, Part
I: Hyperthermic Effects in Man and Spon-
taneous Animal Tumors (to be published).
[3] Pettigrew, R. T., Gait, J. M. , Ludgate,
CM., Horn, D. B., and Smith, A. N.,
Br. J. Surg. 61, 727 (1974).
[4] Greenleaf, J. F. and Johnson, S. A.,
Algebraic Reconstruction of Spatial Dis-
tributions of Acoustic Velocity and At-
tenuation in Tissue from Time-of-Fl ight
and Amplitude Profiles, in Ultrasonic Tis-
sue Characterization, M. Linzer, ed..
National Bureau of Standards Spec. Publ.
453, pp. 109-119 (U.S. Government Printing
Office, Washington, D.C., 1976).
60
Rschevkin, S. N., The Theory of Sound,
pp. 440-444 (Macmillan, New York, 1963).
Lees, S., Gerhard, F. B. Jr., and Oppen-
heim, F. G., Ultrasonics 11, 269 (1973).
Carstensen, E. L. and Schwan, H. P., J_.
Acoust. Soc. Amer. 3i, 305 (1959); Carsten-
sten, E. L., Absorption of Sound in Tissue
(this publication, p. 29).
[8] Icsevgi , A. and Lamb, W. E. Jr., Phys . Rev.
185, 517 (1969). Section IV of this paper
gives an elegant and clear explanation of
the behavior of the pulse velocity when the
group velocity exceeds the phase velocity.
[9] Lovett, J. R., J. Acoust. Soc. Amer. 45,
1051 (1969).
61
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed.. National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
ULTRASONIC ATTENUATION IN NORMAL AND ISCHEMIC MYOCARDIUM
M. O'Donnell, J. W. Mimbs, B. E. Sobel , and J. G. Miller
Washington University
St. Louis, Missouri 63130, U.S.A.
The ultrasonic attenuation coefficient of dog myocardium was measured in vitro
over the frequency range 2 to 10 MHz. Changes in the attenuation of normal myocardium
were measured as a function of time after excision at fixed temperatures. Results of
measurements made on tissue maintained at 35 °C revealed progressive changes in the
attenuation as a function of time, presumably indicative of tissue degradation. Re-
sults of measurements at 19.5 °C, however, showed no significant changes in attenuation
up to 4 hours following excision. The temperature dependence of the attenuation was
measured over the range 20 ^C to 37 °C, yielding the result that the attenuation
coefficient at 37 °C is about 20 percent lower than that at 20 °C. The attenuation co-
efficient was measured in vitro at 20 °C in hearts from dogs previously subjected to
coronary occlusion and sacrificed at intervals ranging from 15 minutes to 3 days fol-
lowing occlusion. Results of these measurements indicate a modest decrease in the
attenuation of ischemic tissue measured at 15 minutes, 1 hour, 6 hours and 24 hours,
and an increase in attenuation of ischemic regions studied 3 days following occlusion.
Key words: Ischemic injury; myocardial infarction; ultrasonic attenuation.
1. Introduction
Previous reports from our laboratory presented
the ultrasonic attenuation in normal myocardium
and in myocardium subjected to ischemic injury
from dogs sacrificed 4 to 11 weeks after coronary
artery occlusion. In this report, we present the
results of a new series of attenuation measure-
ments on dog left ventricle. Results from this
series of measurements are analyzed to: (1)
contrast attenuation in normal and ischemic zones
of tissue from animals sacrificed 15 minutes,
1 hour, 6 hours, 24 hours, and 3 days following
coronary occlusion, (2) examine changes in the
ultrasonic attenuation of tissue in vitro at fixed
temperatures as a function of time after excision,
and (3) examine the temperature dependence of the
ultrasonic attenuation of tissue in vitro.
2. Methods
A. Ultrasonic analysis
TRANSMITTER
SPECIMEN RECEIVER
DRIVER
TIMING UNIT
SPECTRUM
ANALYZER
GATE
SAMPLE
a HOLD
ADC
ADC
DIGITAL
PROCESSING
STORAGE
Fig. 1. Block diagram of the ultrasonic
instrumentation used in this study.
The instrumentation employed for ultrasonic
analysis is depicted in figure 1. The transmit-
ting transducer and driver yielded an ultrasonic
pulse with frequency components of sufficient
amplitude to permit operation over a range of
2 to 10 MHz. Details of transducer design are
discussed subsequently. Under control of a tim-
ing unit, broadband ultrasonic pulses were gated
into a slowly sweeping analogue spectrum analyzer.
Output pulses from the spectrum analyzer, compris-
ing the logarithm of the Fourier transform of the
received ultrasonic pulses, were converted by a
sample-and-hold unit into a slowly varying (dc)
signal. As the spectrum analyzer was slowly
swept through frequency, the output of the sample-
and-hold and a voltage corresponding to the fre-
quency being analyzed by the spectrum analyzer were
transmitted to two analogue-to-digital converters
(ADC's). The digital outputs from the ADC's were
recorded on magnetic tape for subsequent analysis.
The procedure used for the quantitative measure-
ment of the attenuation following transmission of
ultrasound through tissue has been described in
detail in previous reports [1,2]^ and is similar
to a substitution technique developed by Schwan
^Figures in brackets indicate literature
references at the end of this paper.
53
and Carstensen [3]. If reflections at saline-
tissue interfaces can be neglected (and in the
absence of artifacts arising from phase cancella-
tion effects), the ultrasonic attenuation coeffi-
cient of tissue of thickness Az is given by
[Va(v) - Vb(v)] , .
at = ; ^ '
^ AZ log^oe
where the term [V/\(v) - Vb(v)] represents the
signal loss, which is obtained by subtracting the
logarithmic output of the spectrum analyzer in
the presence of tissue [Vd(v)] from that recorded
in the absence of tissue IVa(v)]. The signal loss
(expressed in decibels) is calibrated by comparison
with an electronic attenuation standard.
Figure 2 illustrates the use of the method to
determine the attenuation coefficient of a test
object, a flat polyethylene plate of Az = 0.5 cm.
The left panel of figure 2 exhibits a plot of
Va(v) with no specimen present (upper curve) and
a plot of Vb(v) with the polyethylene specimen
present (lower curve). The result of subtraction
of the two curves in the left panel of figure 2 is
displayed as the attenuation-frequency plot in the
right panel, with data points calculated according
to eq. (1). For cases in which reflections from
saline-specimen interfaces cannot be neglected,
the slope of the attenuation coefficient versus
frequency curve displayed in figure 2 remains a
valid ultrasonic index, although the attenuation
coefficient may not. (The ultrasonic pulse
length used in these studies was less than 0.05 cm
in a polyethylene plate of Az = 0.5 cm. Thus
standing wave effects could not occur.)
A possible source of artifact in the apparent
attenuation coefficient as determined from eq.
(1) arises from phase cancellation effects [1,2].
Phase cancellation effects may occur if inhomo-
geneities in the tissue distort the ultrasonic
field presented to a spatially extended piezo-
electric receiving transducer. These wavefront
distortions may result from transmission of ultra-
sound through tissue with variations in surface
characteristics, internal structural characteris-
tics, or both. When the wavefronts incident upon
a piezoelectric receiver are distorted, the
generated electrical signal is degraded because
of the phase sensitive nature of the receiver.
Phase cancellation effects can induce artifacts
into attenuation data which might be interpreted
incorrectly as reflecting only the absorption and
scattering properties of a specimen [1,2,4].
An intensity sensitive ultrasonic receiving
transducer which is inherently insensitive to
phase cancellation effects is under continuing
development in our laboratory [1,2]. This trans-
ducer makes use of the acoustoelectric effect in
single crystal cadmium sulfide to convert the in-
tensity of an incident ultrasonic wave into an
electronic current in the semiconducting crystal.
An illustration of phase cancellation effects is
presented in figure 3, which contrasts the per-
formance of a piezoelectric and an acoustoelectric
transducer of equal diameters (1.3 cm). The upper
panel of figure 3 illustrates the apparent attenu-
ation coefficient as a function of frequency for
four adjacent sites of dog left ventricle mea-
sured with the piezoelectric receiver. Marked
variability is observed in data obtained from four
morphologically similar regions. The lower panel
depicts the results of measurements on the identi-
cal four sites; however, the intensity sensitive
acoustoelectric receiver was used. The data of
the lower panel exhibit more uniform consistency,
presumably reflecting the reduction or elimination
of artifacts due to phase cancellation.
The present configuration of the acousto-
electric receiver is cumbersome and would not
permit the rapid collection of data that was re-
quired in experiments on ischemic tissue (see
below). All data reported here were obtained
using identical focused piezoelectric transducers
(1.3 cm diameter, 5 cm focal distance) for trans-
mitter and receiver. By positioning the specimen
of interest so that it lay entirely within the
depth of field of the focused transducers, phase
cancellation artifacts were minimized. A com-
parison of the performance of the focused trans-
ducers and the acoustoelectric receiver is shown
in figure 4. In this figure the apparent at-
tenuation coefficient versus frequency plots ob-
tained from analysis of the identical four ad-
jacent sites of left ventricle are presented.
The upper panel presents data obtained with the
Fig. 2. The left panel is a plot of Va(v) obtained without a specimen and Vb(v) obtained
with a specimen (a polyethylene plate of Az = 0.5 cm in this case). The right
panel depicts the attenuation versus frequency plot determined using eq. (1) and
the data shown in the left panel.
64
0.8
0.6
0.4
0.2-
0.0
1.0
0.8 -
0.6 -
0.6
0.4
Piezoelectric receiver
Acoustoel ectric receiver
O.OL
4 6
FREQUENCY (MHz)
Fig. 3. Planar piezoelectric transmitter. The
upper panel depicts the apparent attenu-
ation coefficient as a function of fre-
quency for four contiguous sites of dog
left ventricle measured with a 1.3 cm
diameter planar piezoelectric receiver.
The lower panel depicts the results of
measurements on the identical four sites
using a 1.3 cm diameter intensity sensi-
tive acoustoel ectric receiver.
0.8r
0.6
0.4
0.2
0.0
0.8
0.6
0.4
0.2
Focused piezoelectric
receiver
0.0
Acoustoelectric receiver
4 6
FREQUENCY (MHz)
Fig. 4. Focused piezoelectric transmitter. The
upper panel illustrates the apparent at-
tenuation coefficient as a function of
frequency for four contiguous sites of dog
left ventricle measured using a 1.3 cm di-
ameter focused (5 cm focal distance) piez-
oelectric receiver. The lower panel de-
picts the results of measurements on the
identical four sites, using a 1.3 cm diam-
eter acoustoelectric receiver.
focused piezoelectric receiver and the lower
panel presents data obtained with the acousto-
electric receiver. A focused piezoelectric trans-
mitter was used to obtain the results shown in
both panels of figure 4. The focused piezo-
electric transducer arrangement (fig. 4) is seen
to offer significant improvement over the planar
piezoelectric transducer arrangement of figure 3.
Data reproducibility and consistency remain,
however, more satisfactory with the use of the
acoustoelectric receiver.
In an attempt to quantitate the extent to
which phase-cancellation effects resulted in de-
gradation of attenuation-frequency data, a
statistical index was developed. Because phase
cancellation effects in attenuation-frequency
data may manifest themselves as erratic (non-
monotonic) frequency dependences (in contrast with
the expected monotonic variation of attenuation
with frequency), one method of segregating the
data is on the basis of a statistical test of the
goodness of a fit to a monotonic curve. Using a
65
three parameter fit to the measured signal loss
date [(V/\-Vb) of eq. (1)], a root mean square
deviation (RMSD) was calculated. The RMSD was
defined by
RMSD = (v, - "0'' jV [ym(-^') - yc(v)] j dv (2)
where v is the ultrasonic frequency, - v-^) de-
fining the frequency interval of the measurements.
y^{v) is the measured signal loss at frequency v,
and y(-(v) is the calculated signal, loss, deter-
mined by a least squares three parameter fit to
the data, ihe units of RMSD are the same as those
of y(v), in this case decibels (dB). The value of
RMSD was used to quantitate the degree of non-
monotonicity and thus to estimate in a relative
fashion the contribution of phase cancellation ef-
fects. Large values of RMSD would be expected
whenever substantial phase cancellation contribu-
tions influenced the apparent attenuation.
The application of this approach is demonstrat-
ed in figure 5, which illustrates our progress in
systematically reducing artifacts due to phase
cancellation. The calculated RMSD for each at-
tenuation-frequency curve obtained with several
transducer combinations is shown. Data obtained
in 20 measurements with a 1.3 cm diameter planar
piezoelectric receiver are shown in panel A and
in 44 measurements using a 0.2 cm diameter planar
piezoelectric receiver in panel B. In both cases
the transmitter was a 1.3 cm diameter planar
piezoelectric transducer. The reduced RMSD's of
1 3cm diameter A
1 piezoelectric receiver
N= 20
/
\ 0 2 cm diameter n
piezoelectric receiver
1 N -- 44
1
1 Focused piezoeleclric r-
, receiver and tionsmitter
1 N = 42
1 1.3cm diameter
' acoustoelertf ic receiver D
1 N : 33
1 ill 1 1 1 1 1 1
00 05 10 15 20 2.5 30 35 4 0 45
ROOT MEAN SQUARE DEVIATION (dB)
Fig. 5. Histogram showing the range of cal-
culated root mean square deviations
(RMSD's) of attenuation-frequency
curves determined with four
different transducer configurations.
panel B as compared with panel A presumably re-
flect a decrease in the phase cancellation arti-
facts achieved by reducing the area over which
the ultrasonic field was integrated. Panel C
exhibits the results of measurements on 42 sites
made with a focused transducer pair and demon-
strates substantial improvement over the arrange-
ments of panels A and B. (It should be noted,
however, that substantial artifacts due to phase
cancellation occur in the presence of intervening
tissue located outside of the focal zone. Thus,
for future applications in vivo, the use of a
focused transducer does not appear to be a satis-
factory method for minimizing artifacts due to
phase cancellation.) The results of the measure-
ments on 33 sites using the acoustoelectric re-
ceiver are presented in panel D of figure 5. The
very small RMSD's obtained with the intensity
sensitive acoustoelectric receiver reflect the
inherent freedom from phase cancellation arti-
facts, making a transducer of this design poten-
tially suitable for future in vivo applications.
B. Tissue preparation and independent
indices of pathology
All measurements were carried out in vitro on
myocardial tissue obtained from adult, mongrel,
15 to 30 kg dogs. In studies of ischemic injury,
each dog was intubated after anesthesia with
sodium pentobarbital (25 mg/kg, intravenously),
placed on a Harvard respirator, ventilated with
room air, and subjected to a left thoracotomy via
the fifth interspace. The pericardium was in-
cised, and the left anterior descending coronary
artery dissected free immediately distal to the
first ventricular branch and ligated. The peri-
cardium was left open. For dogs studied at 6
hours, 24 hours, and 3 days following occlusion,
the chest was closed conventionally, with intra-
pleural suction maintained for at least one hour
via a chest tube.
At the time of sacrifice, each animal was again
anesthetized with sodium pentobarbital (50 mg/kg,
intravenously). The heart was rapidly excised
and placed in a 0.9 percent NaCl solution. To
prepare the tissue for ultrasonic analysis an in-
cision was made at the root of the aorta, con-
tinued interiorly along the left ventricular sur-
face of the interventricular septum to the apex,
and extended posteriorly and superiorly, terminat-
ing at the base of the heart. This permitted
prompt excision of a segment of the left ventric-
ular anterior and apical wall. This segment com-
prised the area of infarction and surrounding
area of normal myocardium in all animals subject-
ed to coronary ligation. (We note that this seg-
ment was measured unaltered, i.e., no additional
preparation such as further slicing to achieve
flat or parallel surfaces was undertaken.) Except
as noted, ultrasonic analysis was initiated within
5 minutes following the death of the animal, and
completed within 45 minutes.
Tissue from animals sacrificed at 24 hours and
72 hours following coronary occlusion was analyzed
biochemically for creatine kinase (CK) content,
an established index of tissue pathology. One
centimeter diameter biopsies corresponding to re-
gions of ultrasonic analysis were obtained im-
mediately following ultrasonic measurement. To
facilitate comparison of data from different ani-
mals, regional CK activity was expressed as the
percentage of depletion compared to activity in
normal myocardium from the same animal. Sites
with CK depletion of greater than 40 percent were
classified as ischemic and those with CK depletion
less than 20 percent were classified as normal.
For dogs sacrificed from 15 minutes to 6 hours
following occlusion, colloidal carbon black was
66
injected into the left atrium fifteen seconds
prior to killing the animal for the purpose of
differentiating regions of ischemic and normal
myocardium. Regions of ischemia did not change
color after injection, while non-ischemic regions
rapidly were stained black. Visual inspection of
the tissue after excision was used to differen-
tiate normal from ischemic tissue. Analysis of
60 sites from 4 control dogs indicated that the
presence of dye did not alter the ultrasonic
measurements of normal myocardium.
3. Results
Prior to addressing questions related to
ischemic injury, it is necessary to determine ex-
perimentally the range of variability introduced
into the results of ultrasonic attenuation mea-
surements in vitro by variables such as: (1) the
time interval that elapses between the sacrifice
of the dog and the ultrasonic measurement, (2)
the temperature of the tissue during ultrasonic
measurement, and (3) the specific region of the
left ventricle that is analyzed.
A. Time interval between sacrifice
and measurement
In an effort to estimate the extent to which
tissue degradation compromises the results of in
vitro measurements, the ultrasonic attenuation
coefficient of dog myocardium was measured at
Time
after
exci sii
15 min
2 h
4 h
15 min
2 h
4 h
B. Temperature dependence of the attenuation
To relate the results obtained at 20 °C to
what might be expected if the measurements were
carried out at 37 °C, an additional series of
experiments was designed to elucidate the tem-
perature dependence of the attenuation in a
manner relatively independent of effects aris-
ing from tissue degradation. Results from the
previous series of experiments (table 1) sug-
gested the possibility that tissue might be
temporarily stored at approximately 20 °C
without undergoing significant degradation be-
tween measurements made at elevated temperatures.
Using this approach, four freshly excised hearts
were used for a measurement of the attenuation at
20.5 °C, 25 °C, 30 °C, and 37 °C. Each segment
of left ventricle was placed in a 20.5 °C bath
and analyzed ul trasoni cally immediately after
constant temperature as a function of time fol-
lowing excision. All experiments were performed
in a 0.9 percent NaCl bath with temperature re-
gulated to ± 0.2 °C by means of a controller and
reci rcul ator.
In table 1 data are displayed as a function of
time following excision for experiments performed
at two temperatures: (1) 19.5 °C (analysis of 27
regions from 3 hearts) and (2) 35 °C (analysis of
27 regions of 3 additional hearts). The ultra-
sonic attenuation coefficient as a fraction of
frequency was approximately independent of time
up to 4 hours following excision for tissue main-
tained at 19.5 °C. When the temperature was
maintained at 35 °C, however, a definite change
with time was observed. For measurements at
35 °C, the value of the slope of a least squares
line fit to the attenuation versus frequency data
changed from (0.061 ± 0.003) cm-iMHz-i (mean ±
standard error (S.E.)) at 15 minutes following
sacrifice to (0.071 + 0.003) cm-iMHz-i at 4
hours following sacrifice. It thus appears that
no statistically significant changes in attenua-
tion occur for periods up to several hours fol-
lowing excision for tissue maintained at approxi-
mately 20 °C, although small {■'^ 15 percent) but
significant changes occur over that time interval
for tissue maintained at 35 °C. These results
motivated the choice of 20 °C as the temperature
at which the extensive series of measurements on
ischemic and normal myocardial tissue in vitro
reported below were carried out.
excision. After analysis at this temperature,
the tissue sample and holder were removed from
the measurement bath and placed in a storage bath
which was maintained at 19.5 °C. The measurement
bath was heated to 25 °C and regulated to maintain
this temperature to ± 0.2 °C. Upon stabilization
of the temperature at 25 °C, the sample and hold-
er were reinserted into the original bath and al-
lowed to equilibrate, whereupon the ultrasonic
attenuation was measured. The same procedure was
repeated for measurements at 30 °C and 37 °C,
with the entire process lasting about 80 minutes.
To check the validity of this method for identify-
ing the true temperature dependence, the attenua-
tion obtained at 37 °C using the above procedure
was compared to the attenuation measured at 37 °C
for 3 additional freshly excised hearts at 27
sites. The slope of the attenuation for these
measurements made at 37 °C within minutes of ex-
Table 1. Ultrasonic attenuation of normal myocardium as a function of time after excision.
Number Number Temp. Attenuation coefficient (cm"') (mean + SE) Slope of a vs^
of of ■ frequency
sites
dogs
O
C
2
MHz
4
MHz
6
MHz
8
MHz
10
MHz
(
cm '
MHz
-1)
27
3
19
.5
0.
10
± 0
02
0.
19
± 0
02
0
34
± 0
02
0
38
± 0.03
0
65
± 0
03
0
072
± 0
002
27
3
19
.5
0
10
± 0
02
0.
20
± 0
02
0
36
± 0
02
0
50
± 0.03
0
70
z 0
03
0
075
± 0
002
27
3
19
.5
0
10
± 0
02
0
20
± 0
02
0
35
± 0
02
0
49
± 0.03
0
69
± 0
03
0
.075
± 0
002
27
3
35
0
09
+ 0
02
0
17
± 0
03
0
30
± 0
03
0
43
+ 0.03
0
56
± 0
04
0
.061
± 0
003
27
3
35
0
09
± 0
02
0
19
± 0
03
0
34
± 0
03
0
47
± 0.03
0
61
± 0
04
0
.068
± 0
003
27
3
35
0
11
± 0
02
0
22
± 0
.03
0
37
± 0
03
0
52
± 0.03
0
64
± 0
04
0
.071
± 0
003
67
Table 2. Ultrasonic attenuation of normal myocardium as a function of temperature.
Number Number Temp. Attenuation coefficient (cm"^) (mean ± SE) Slope of a vs
of
ites
of
dogs
2
MHz
4
MHz
6
MHz
8
MHz
10
MHz
frequency
(cm-' MHz-i)
36
4
20.5
0
10
± 0
02
0
19
± 0
02
0
34
± 0
03
0
48
+ 0
03
0
64
± 0
03
0.071 + 0.002
36
4
lb
U
10
± 0
02
0
19
+ 0
02
0
33
± 0
03
0
45
± 0
03
0
61
± 0
03
0.058 ± 0.UU2
36
4
30
0
10
± 0
02
0
17
± 0
02
0
31
± 0
03
0
43
± 0
03
0
58
± 0
03
0.054 ± 0.002
36
4
37
0
11
± 0
02
0
19
± 0
02
0
31
± 0
03
0
41
± 0
03
0
56
± 0
03
0.058 ± 0.002
cision was (0.060 ± 0.002) cm"i MNz"! in good
agreement with the value 0.058 ± 0.002 cm-^ MMz"!
measured at 37 °C for the 36 sites of tissue
stored at 19.5 °C between measurements. This re-
sult suggests that the procedure is methodologi-
cally sound.
Results of this series of experiments on 36
regions from 4 hearts designed to determine the
temperature dependence of the attenuation coeffi-
cient are presented in table 2. The slope of a
least squares line fit to the attenuation-
frequency data decreases approximately linearly
with increasing temperature over the range
20.5 °C to 37 °C, with the value at 37 °C being
about 20 percent less than that at 20.5 °C.
C. Regional variations in attenuation
In order to identify possible variations in the
attenuation coefficient as a function of the re-
gion of the left ventricle that was analyzed, 102
sites from 4 dogs not subjected to coronary liga-
tion were studied and the results segregated into
4 groups based on the region investigated. Data
from these experiments are presented in table 3.
Values of the attenuation coefficient for regions
in the mid-posterior and apical aspects of the
left ventricle were essentially identical. Some-
what higher values were exhibited by regions in
the papillary muscles and generally lower values
were exhibited by regions near the base of the
heart. These results are summarized in figure 6,
utilizing the slope of a least squares line fit
to the attenuation coefficient versus frequency
plots for comparative purposes.
Based on the results of this study of regional
variability of the ultrasonic attenuation coeffi-
cient of normal myocardium, subsequent studies
were conducted primarily on sites from mid-
posterior and apex, avoiding sites from the
papillary muscles and the base. With this ap-
proach, the value of the slope of the attenuation
for normal left ventricle, as determined from
measurements at 245 discrete sites from 36 dogs,
was (0.072 ± 0.001) cm"! MHz'i (mean ± S.E.).
0.08
iS 0.06
0.04
LU
o 0.02
posterior
aspect
=21
papi 1 lary
muscl es
7^
//
'//
/'
/ /
f.
N=38
N=21
N=21
apex
mid-
base
posterior
Fig. 6. Regional variation of the slope of the
attenuation (2 to 10 MHz) in normal
myocardium.
Number Number Temp
Descrip- of of
tion sites dogs °C
Table 3. Regional variation of ultrasonic attenuation in normal myocardium.
Attenuation coefficient (cm"M (mean ± SE)
2 MHz
4 MHz
5 MHz
8 MHz
10 MHz
Slope of g vs
frequency
(cm'i MHz-i)
20 0.11 ± 0.02 0.20 ± 0.02 0.34 ± 0.02 0.45 ± 0.03 0.62 + 0.03 0.064 ± 0.002
20 0.10 ± 0.02 0.19 ± 0.02 0.34 ± 0.02 0.48 ± 0.03 0.64 ± 0.03 0.070 ± 0.002
Base
Mid-
posterior
Apex
21
21
39
Papillary 21
muscles
20 0.10 ± 0.02 0.19 ± 0.02 0.35 ± 0.02 0.49 ± 0.03 0.65 ± 0.03 0.071 ± 0.002
20 0.13 + 0.02 0.26 ± 0.02 0.41 ± 0.02 0.55 ± 0.03 0.76 ± 0.03 0.079 ± 0.002
68
D. Effects of ischemic injury
To investigate changes in physical properties of
tissue resulting from ischemic injury, ultrasonic
measurements were carried out on freshly excised
hearts previously subjected to coronary occlusion
for intervals of 15 minutes, 1 hour, 6 hours, 24
hours, and 3 days. Results of these measurements
are summarized in table 4. The attenuation meas-
ured at 15 minutes, 1 hour, 6 hours, and 24 hours
following coronary occlusion is lower than that ex-
hibited by normal myocardium, while that at 3 days
following occlusion is higher than that of normal.
The attenuation values reported in table 4 rep-
resent averages of the values obtained for all
sites analyzed. Results illustrated in figures 4
and 5 suggest that artifacts arising from phase
cancellation effects may occur in data obtained
using the focused transmitter-focused receiver con-
figuration. To eliminate measurements exhibiting
probable phase cancellation effects from further
consideration, an index based on RMDS (eq. 2) was
used to segregate the data. Attenuation coeffi-
cient versus frequency curves exhibing RMSD's
greater than 0.25 dB were deleted from further con-
sideration. The value 0.25 dB was selected for
segregating the data because it represents an RMSD
two times the maximum RMSD exhibited by data ob-
tained using the phase cancellation insensitive
acoustoel ectric receiver. The results of segre-
gating the data in this way are illustrated in
table 5. The average values of the slope of the
attenuation versus frequency curves are presented
for two groupTl {^) data including sites exhibit-
ing phase cancellation effects, and (2) data ex-
cluding sites exhibiting phase cancellation effects.
A comparison of the number of sites contributing
to the average slope in each group indicates that a
modest but significant number of attenuation versus
frequency curves were compromised by phase cancel-
lation effects.
The results of experiments on normal and ische-
mic myocardium are summarized in figure 7. The
average slope of the attenuation (last column of
table 5) is shown at 15 minutes, 1 hour, 6 hours,
24 hours, and 3 days following coronary occlusion.
The slope of the attenuation of ischemic myocardium
is significantly lower than that of normal myocard-
ium for times ranging from 15 minutes through 24
hours following coronary occlusion and is signifi-
cantly larger at 3 days following occlusion.
Descrip- of
tion sites
Table 4. Ultrasonic attenuation of normal and ischemic myocardium
Number Number Temp. Attenuation coefficient (cm'^) (rtiean ± SE)
C
of
dogs
2 MHz
4 MHz
6 MHz
8 MHz
10 MHz
Slope of g vs
frequency
(cm-' MHz-')
Normal
Time after
occl usion
245
36
20
0.10 ± 0.02 0.19 ± 0.02 0.33 ± 0.02 0.48 ± 0.03 0.65 ± 0.03 0.072 ± 0.001
15
min
56
5
20
0
10
0
02
0
17
+
0
02
0
32
+
0
02
0
43
+
0
03
0
57
+
0
03
0
063
+
0
003
1
h
64
8
20
0
09
+
0
02
0
17
0
02
0
32
4-
0
02
0
43
+
0
03
0
57
+
0
03
0
063
+
0
003
6
h
59
5
20
0
08
0
02
0
15
+
0
02
0
30
+
0
02
0
44
0
03
0
57
+
0
03
0
066
+
0
002
24
h
29
11
20
0
08
0
02
0
16
+
0
02
0
29
+
0
02
0
43
0
03
0
59
+
0
03
0
067
+
0
002
3
d
31
7
20
0
11
+
0
02
0
20
+
0
02
0
38
+
0
02
0
52
0
03
0
69
+
0
03
0
079
+
0
002
Table 5. Slope of the ultrasonic attenuation of normal and ischemic myocardium.
Including sites Excluding sites
exhibiting phase exhibiting phase
Description cancellation effects cancellation effects
Time after Numoer Temperature Character Number Slope Number Slope
coronary occlusion of dogs °C of sites of sites (cm"' MHz"M of sites (cm"^ MHz'M
15
min
5
20
normal
58
0
073
±
0
003
45
0
072
+
0
003
15
min
5
20
ischemic
56
0
063
+
0
003
49
0
062
+
0
002
1
h
8
20
normal
69
0
072
+
0
003
52
0
071
0
003
1
h
8
20
i schemic
64
0
063
+
0
003
57
0
062
+
0
003
6
h
5
20
normal
59
0
072
+
0
002
51
0
072
+
0
002
6
h
5
20
ischemic
59
0
066
+
0
002
49
0
064
+
0
002
24
h
11
20
normal
29
0
073
+
0
002
24
0
068
+
0
002
24
h
11
20
ischemic
29
0
067
+
0
002
27
0
060
+
0
002
3
d
7
20
normal
30
0
072
+
0
002
25
0
072
+
0
001
3
d
7
20
i schemic
31
0
079
+
0
002
28
0
082
0
002
£9
I 0 08
M
X
.0 07
So 06
o
005
Fig. 7.
0.09
i
NORMAL
ISCHEMIC
N=
24 27
I5min Ih 6h 24 h
TIME AFTER OCCLUSION
25 28
3d
Slope of the ultrasonic attenuation
(2 to 10 MHz) for normal and ischemic
myocardial tissue determined at specified
intervals following coronary occlusion.
tenuation might be expected to accompany edema,
which is a recognized correlate of recent
ischemic injury. If the small decrease in at-
tenuation of ischemic tissue is corroborated by
measurements ir[ vivo, the use of an index based
on the ultrasonic attenuation to characterize the
state of myocardium might be approached. Efforts
to use an index based on attenuation to charac-
terize the state of myocardial tissue might
utilize each heart as its own control thus abro-
gating data variability due to regional myocardial
variations as documented in table 3.
An increase in attenuation over that of normal
myocardium is observed in tissue subjected to
ischemic injury and measured three days following
coronary occlusion. Previous reports from this
laboratory documented substantial increases in at-
tenuation in tissue studied 4 to 11 weeks after
coronary occlusion [2,9]. This increase in at-
tenuation, evident at three days following occlu-
sion and reaching substantial magnitude weeks
after occlusion, appears to be related to the on-
set of scar formation, reflecting the increase in
the collagen content of necrotic tissue.
4. Discussion
Emphasis in the present studies was placed on
systematic measurements making use of carefully
defined control experiments. Either creatine
kinase depletion or reduced uptake of colloidal
carbon was used as an independent index for
documenting myocardial ischemia. Care was taken
to eliminate artifacts due to ultrasonic phase
cancellation effects, a potentially significant
source of variability in attenuation coefficient
measurements. Methods suitable for conducting
meaningful experiments on excised tissue were de-
veloped and validated. Results of the study
addressing the problem of tissue degradation fol-
lowing excision led to the choice of ^ 20 °C
rather than 37 °C for the present series of in-
vestigations. The possibility of significant
changes in the ultrasonic properties of tissue
measured in vitro as a function of time follow-
ing excision was considered by Hueter [5], and
more recently by Frizell and Carstensen [6], and
by Bamber et al . [7]. Previous studies examined
changes for time intervals substantially larger
(tens of hours) than the present experiments,
which had the more limited goal of defining the
range of validity of attenuation measurements on
freshly excised tissue.
The temperature dependence of the attenuation
of normal myocardium was measured over the range
^. 20 °C to 37 °C using a technique designed to
minimize artifacts arising from tissue degrada-
tion. Results of these measurements suggest that
the data obtained at 37 °C might be expected to
differ by only about 20 percent from that ob-
tained at 20 °C. However, these limited measure-
ments on normal myocardium do not adequately ad-
dress the possibility of differences in the tem-
perature dependence of the attenuation exhibited
by ischemic as opposed to normal tissue.
Results of this study indicate a small de-
crease in the attenuation in the early stages of
a myocardial infarct (up to 24 hours following
coronary occlusion). Although mechanisms re-
sponsible for the attenuation of soft tissue are
inadequately understood [8], a decrease in at-
Acknowl edgments
Lawrence J. Busse carried out the investiga-
tions of phase cancellation effects described in
the text and presented in figures 3, 4, and 5.
Pranoat Suntharothok-Priesmeyer was responsible
for production of the text and illustrations.
This work was supported in part by grants
HL19537, HL17646, and HL07081 from the National
Institutes of Health.
References
[1] Busse, L. J., Miller, J. G., Yuhas, D. E.,
Mimbs, J. W., Weiss, A. N., and Sobel , B. E.,
Phase Cancellation Effects: A Source of
Attenuation Artifacts Eliminated by a CdS
Acoustoelectric Receiver, in Ul trasound
in Medicine, D. White, ed., vol. 3, pp.
1519-1535 (Plenum Press, New York, 1977).
[2] Miller, J. G., Yuhas, D. E., Mimbs, J. W. ,
Dierker, S. B., Busse, L. J., Laterra, J. J.,
Weiss, A. N., and Sobel, B. E., Ultrasonic
Tissue Characterization: Correlation Between
Biochemical and Ultrasonic Indices of Myo-
cardial Injury, in Proceedings 1976 IEEE
Ultrasonics Symposium 79, 33-43 (Cat. No.
CM 1120-43U, IEEE, New York, 1976).
[3] Schwan, H. P. and Carstensen, E. L., Ultra-
sonics aids in diathermy experiments.
Electronics, 216-220 (July 1952).
[4] Marcus, P. N. and Carstensen, E. L., Problem
with absorption methods of inhomogeneous
solids, J. Acoust. Soc. Am. 58_, 1334-1335
(1975).
[5] Hueter, T.A., WADC Tech. Rept. 57-706 (1958).
[6] Frizell, L. A., Ultrasonic Heating of Tissue,
Ph.D. Thesis, University of Rochester (1976)
(unpublished).
70
[7] Bamber, J. C, Fry, M. J., Hill, C. R. , and
, Dunn, F., Ultrasonic attenuation and back-
scattering by mammalian organs as a function
of time after excision, Ultrasound in Medi-
cine and Biology 3, 15-20 TT977T-
[8] See the references listed in O'Donnell, M.
and Miller, J. G. , Mechanisms of Ultrasonic
Attenuation in Soft Tissue (this publication,
p. 37).
[9] Yuhas, D. E. , Mimbs, J. W., Miller, J. G.,
Weiss, A. N., and Sobel , B. E., Changes in
Ultrasonic Attenuation Indicative of Regional
Myocardial Infarction, in Ultrasound in
Medicine, D. W. White, ed., vol. 3, 1883-
1894 (Plenum Press, New York, 1977).
71
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed., National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
ACOUSTIC MICROSCOPIC ANALYSIS OF MYOCARDIUM
Donald E. Yuhas and Lawrence W. Kessler
Sonoscan Inc.
720 Foster Avenue
Bensenville, niinois 60106, U.S.A.
Acoustic microscopy can be employed to measure variations in the ultrasonic attenua-
tion and velocity over spatial dimension of tens of micrometers. Knowledge of the in-
teractions at this fine-structure level provides a more complete understanding of the
response of tissues at diagnostic frequencies. In this article we investigate the ultra-
sonic characteristics of formalin-fixed infarcted and normal myocardium at the micro-
scopic level. The elastic microstructure of normal tissues is found to be uniform over
spatial dimensions of hundreds of micrometers. The attenuation coefficient a, at
100 MHz, ranges from 37 to 68 cm"^ while the velocity varies less than ± 20 m/s.
Acoustic anisotropy, attributed to muscle fiber orientation, is found to be a primary
source of attenuation variability. The elastic microstructure in the zones of infarc-
tion is very distinct from that in normal zones. With infarction, variations in the
attenuation ranging from 38 to 100 cm"i occur over dimensions of hundreds of micro-
meters. Accompanying this variability in attenuation is a wide distribution of velocity.
The most highly attenuating regions in the infarcted tissue show localized increases in
the velocity of more than 75 m/s. Such large variations in sonic velocity are sufficient
in magnitude to be an important source of phase cancellation in low frequency attenua-
tion measurements. Additionally, the 100 MHz data indicate that the frequency depend-
ence of the attenuation coefficient observed in the 1 to 10 MHz range for infarcted and
normal myocardium may not hold in the 10 to 100 MHz frequency interval.
Key words: Acoustic microscopy; anisotropy; attenuation; elastic microstructure;
infarct; interferogram; phase cancellation; myocardium; velocity.
1. Introduction
The complete ultrasonic characterization of
tissue can be divided into two distinct levels
of interaction: 1) structures greater than the
wavelength of diagnostic ultrasound and 2) struc-
tures that are finer. The larger structures
give rise to specular reflections which are used
to define the outlines of tissues and organs
seen in conventional B-scan echograms. The
finer structures give rise to characteristic
scattering properties of tissue and are also
responsible for intrinsic attenuation and velo-
city. It is this sensitivity to fine structure,
that makes the diagnostic frequency ultrasonic
parameters so important in determining tissue
pathologies.
As an aid to understanding the low frequency
scattering, attenuation, and velocity data,
direct visualization of the finer structures is
desirable. Conventional microscopy (optical and
electron) can be used to delineate various struc-
tural components, but these methods do not provide
the required information regarding the ultrasonic
properties of the tissue.
On the other hand, the elastic microstructural
information obtained through acoustic microscopy
is directly applicable to the lower frequency (1
to 10 MHz) characteristics. Acoustic microscopy
provides a means for visualizing directly the
elastic architecture as well as for measuring
variations in the ultrasonic attenuation and
velocity properties over areas tens of micro-
meters in diameter. By using this technique,
valuable insight can be gained into mechanisms
responsible for the attenuation of sound as well
as factors which influence tissue characteriza-
tion measurements at lower frequencies.
This preliminary study consists of a series of
acoustic microscope observations of myocardial
tissue. Previous reports have shown that quanti-
tative changes in ultrasonic attenuation over the
frequency range 2 to 10 MHz are associated with
regional myocardial infarction [1-4]^. The ultra-
sonic attenuation in normal tissue was found to
exhibit a linear frequency dependence, while in-
farcted tissue exhibits a significant quadratic
dependence. As a clue to elucidating the nature
of the change in attenuation at lower frequencies,
acoustic microscopy was used to investigate the
elastic microstructure of normal and infarcted
myocardi urn.
2. Methods
Acoustic micrographs were obtained with a com-
mercially available 100 MHz scanning laser acoustic
^Figures in brackets indicate literature
references at the end of this paper.
73
microscope, SONOMICROSrOPE 100. A brief descrip-
tion of the technique is presented here for com-
pleteness. For a more detailed description, the
reader may refer to the listed references [5-9].
Figure 1 illustrates the instrumentation used
to produce 100 MHz acoustic micrographs. A sample
is placed on a glass stage between an insonifying
transducer and receiver. The transmitting trans-
ducer is a piezoelectric element which is bonded
to a glass substrate, and the receiver is a scan-
ning focused laser beam. The acoustic energy is
coupled to the specimen through the glass sub-
strate without need for an acoustically lossy
water bath. The specimen is moistened with liquid
to ensure good acoustic coupling. The sound is
transmitted through the sample at an angle of 10°
with respect to the normal. In order to provide
a specularly reflecting surface for the scanning
laser beam receiver, the specimen is covered with
a partially silvered coverslip. The coverslip al-
lows a small amount of light to penetrate to the
sample and render an optical view simultaneously
with the acoustic.
laser beam scanners
imaging optics
demodulator and
photodetector
mirrored cover'sl in i g_
spec \men
acoustic
signal processor
acoustic
frequency generator
T
stage
photodetector
ultrasonic transducer
optical
signal processor
Acoustic Image
Display
Fig. 1. Schematic diagram of the acoustic micro-
scope used in this study. The instrument
employs a scanning laser beam technique
for detecting the amplitude and phase of
acoustic energy transmitted through the
sample. Acoustic and optical images are
produced simultaneously and are presented
on separate monitors.
The acoustic energy transmitted through the
specimen imparts a slight oscillatory mechanical
perturbation to the coverslip surface. The ampli-
tude of these perturbations, which vary at the
acoustic frequency, is inversely proportional to
the localized ultrasonic attenuation properties in
the underlying sample. By sensing the perturba-
tions and displaying the signal on a TV monitor,
an acoustic image is formed. Images are obtained
in real time and displayed at a magnification of
about 70 X. The bright regions seen on the
acoustic micrographs correspond to areas of high
transmission through the sample, whereas the
darker areas correspond to regions of higher ultra-
sonic attenuation.
Quantitative attenuation measurements of the
tissue observed in the microscope can be made any-
where within the field of view by an optical tech-
nique described previously [10]. With the sample
in place, an acoustic image is obtained. Areas of
interest are noted and readings of the image bright-
ness are obtained with a light meter. Once the
image brightness is recorded the sample is removed
and replaced by an equivalent thickness of coupling
medium. For this work formalin (10 percent formal-
dehyde) was used. Electrical attenuation is then
inserted into the electro-acoustic signal path to
restore the image brightness to its previous level.
Although the acoustic field is quite uniform,
care was taken to ensure that the same region of
sound field was measured for the cases with and
without the sample in place. Using this technique,
attenuation can be measured over an area as small
as 5 X 10"^ cm2. The sensitivity of the photo cell
used in these studies permitted discrimination of
sound levels to within ± 1 dB, which introduced a
3 to 8 percent uncertainty in the measured at-
tenuation coefficient. Because the acoustic image
of the specimen is observed during the course of
the measurement procedure, it is possible to select
regions of low contrast structural details, there-
by assuring the attenuation values obtained were
free from artifacts arising from vessels or fluid
filled cavities.
In addition to recording the transmissivity of
specimens, the scanning laser acoustic microscope
is equipped with an acoustic interferogram mode of
operation. Acoustic interferograms display the
localized distribution of transit times through
the sample. For specimens of uniform thickness
these transit time variations are directly related
to localized variations in sonic velocity. Inter-
ferograms can be conceptualized with reference to
figure 2. In this schematic, an ultrasonic plane
wave is incident on the sample at an angle with
respect to the normal. The intersection of the
wavefronts with the plane below the specimen re-
sults in a series of fringes (indicated by the
dots) which run in and out of the plane of the
page. Using phase sensitive detection, the posi-
tions of these equiphase lines are displayed on
elastic plane of
inclusion / observation
Fig. 2. Model for conceptualizing acoustic inter-
ferograms. As a plane wave propagates
through a sample, regions of velocity
variations cause refraction. The result-
ing wavefronts are displaced as they ar-
rive at the sample surface as shown by
the dots. The lateral position of the
wavefronts (fringes) are a function of the
change in velocity and can be calculated
from the formulas given in the text.
74
the image monitor. Figure 3 is a micrograph of an
interferogram recorded with no specimen on the
stage. In this case the interferogram lines
(fringes) are straight and parallel. Going from
left to right, each successive interference fringe
represents an increase in transit time through the
glass stage of exactly one period of oscillation
(10 nanoseconds at 100 MHz).
Fig. 3. Interferogram obtained eith no sample on
the microscope stage showing equally
spaced, parallel, interference fringes.
This is the same type of interferogram
that would be obtained on a sample that
exhibited no regional changes in sonic
veloci ty.
Figure 2 also illustrates the effect of elastic
inhomogeneity on the position of the interferogram
fringes. As the wave enters the sample it refracts
according to Snell's law. In the areas of the
sample with uniform propagation velocity (white
areas) the separation of successive fringes after
propagating through the specimen is the same as
that measured with no sample. When structures
with variations in propagation velocity are present,
illustrated by the hatched area, the separation of
successive fringes in the vicinity of the elastic
inclusion changes. The magnitude of the fringe
shift, N, is directly related to the propagation
velocities of the two components by the following
relations [11]:
M = (cot ei - cot 62) (1)
where N is the dimensionless magnitude of the
lateral fringe shift normalized by the unperturbed
fringe spacing: Xi is the wavelength of sound in
region 1 of the specimen; At is the specimen
thickness; and ei and 62 are the propagation angles
in regions 1 and 2 respectively.
The relationship between the propagation angles
and the sonic velocity is given by Snell's law:
sin 9i _ sin 62 ^2)
Ci C2
where Cj and C2 are the respective propagation
velocities.
Using eqs. (1) and (2) and knowing the magni-
tude of the sonic velocity of one component in the
field of view, the sonic velocities of all other
areas can be determined. The primary factor de-
termining the sensitivity of the technique for
measuring variations in velocity is the sample
thickness. For the sample thicknesses used in
this study, a displacement of a fringe by 1/10 of
the unperturbed fringe spacing (i.e., N = 0.1)
corresponds to a velocity change of 5 m/s.
3. Myocardial Samples
The specimens used in this study^ were taken
from regions of formalin-fixed canine myocardium.
Sections .05 ± .005 cm in thickness were made in
several different myocardial regions extending
from the apex to the base of the heart. The plane
of dissection was oriented approximately at an
angle of 45° with respect to the long axis of the
heart and each section extended from epicardial to
endocardial surface. Samples were taken from
regions of infarcted and non-infarcted tissues
with the regions of infarction lying nearer to the
apex of the heart. Analysis was carried out on
section taken from a single heart which had been
formalin fixed for more than a year and was
originally excised 4 weeks after coronary occlusion.
4. Results
Figure 4 presents an acoustic micrograph (a)
and interferogram (b) obtained on a myocardial
section taken from a non-infarcted region. Bright-
ness variations in this micrograph (fig. 4a) gen-
erally range from light to dark grey. A band of
lighter material approximately 1 mm wide can be
seen running almost vertically through the central
portion of the micrograph. On either side of this
acoustically transmissive region are darker areas
oriented both perpendicular and parallel to the
band. Except for the dark structure (indicated by
the arrow), which is attributed to a vessel, the
change in brightness between the various regions
is relatively subdued and gradual. The interfero-
gram of this region (fig. 4b) shows a uniform
sonic velocity distribution. Only the area in-
dicated by the arrow exhibits an abrupt disruption
of the interferogram lines showing a slight in-
crease of sonic velocity of 20 m/s. Throughout
most of the sample the lateral shifts in the inter-
ferogram lines going from the top to the bottom of
the micrograph over the entire field are less than
.4 fringes indicating velocity changes of less
than ± 10 m/s in the vertical direction.
Figure 5 illustrates an interesting effect,
that of acoustic anisotropy in myocardial tissue.
These micrographs were obtained on the same sample
shown in figure 4; however, the sample has been
rotated 90° in the plane of the stage. Recall
that the angle of insonif ication is approximately
10° from the specimen normal. Thus, while the
major component of the propagation vector K is
perpendicular to the sample, a small off-axis com-
ponent leads to some dramatic changes in image
contrast on rotation. The amplitude micrograph
shown in figure 5a reveals two distinct regions.
The upper half of the micrograph shows acoustic
morphology similar to that observed in the pre-
vious micrographs. The lower half of the micro-
^Samples provided courtesy of Washington
University, Division of Cardiology, St.
Louis, Missouri.
75
(a)
Fig. 4. Acoustic amplitude micrographs (a) and
interferogram (b) of normal myocardium.
The field of view in each micrograph is
2.3 by 3.0 mm and the sample thickness is
500 micrometers. In the amplitude micro-
graph the light areas are regions of low
attenuation. The interferograms are ar-
ranged so that a lateral shift to the right
corresponds to a region of increased velo-
city and a shift to the left, lower velo-
city. A shift of one fringe corresponds
to a velocity change of 50 m/s. Typical
variations in this sample are + 10 m/s.
Fig. 5. Acoustic amplitude micrograph (a) and
interferogram (b) of the same region de-
picted in figure 4 after sample rotation
(counter-clockwise) by 90°. The dramatic
changes in these images compared to figure
4 are due to the tissue being anisotropic
acoustical ly.
The top portion of the interferogram shows
velocity variations of ± 5 m/s, while
variations of 40 m/s are seen in the lower
portion of the micrograph.
graph is significantly different, showing bands of
acoustically light and dark material running hori-
zontally across the field of view. The change of
acoustic contrast is abrupt across these bounda-
ries and their morphology suggests that they arise
from muscle fiber orientation. In this micrograph
the off-axis component of the propagation vector
k is parallel to the acoustic striations seen in
the micrograph, i.e. with the grain, whereas in
the previous figure it is perpendicular or across
the grain.
It should be emphasized here that the portion
of the sample comprising the lower half of this
field of view (fig. 5a) is the same area as that
imaged in the left half of figure 4a. Some of
the most highly attenuating areas seen in figure
5a become the least attenuating areas as the
sample is rotated in the sound field. The acoustic
interferogram of the same region is shown in figure
5b. Again, the spatial distribution character of
sonic velocity in the upper portion of the micro-
graph is similar to that seen in figure 4b. The
sonic velocity in this region is 1580 m/s with
typical variations less than ± 5 m/s. In the
darker regions the interferogram lines become
quite jagged, indicative of sharply localized
changes in sonic velocity. Typical shifts here
are on the order of 0.8 normalized fringes, cor-
responding to an increase in velocity of 40 m/s.
Figures 6a and 6b show a comparison between
interferograms obtained on normal and infarcted
tissue. The interferogram obtained on the normal
76
(b)
Fig. 6. Interferograms comparing normal and in-
farcted tissue, (a) Interferogram of
normal tissue shows uniform attenuation
and velocity. Velocity variations are
typically ± 5 m/s. (b) Interferogram of
infarcted tissue. Both attenuation and
sonic velocity are highly variable.
Typically, the highly attenuating areas
show increased sonic velocity, for example,
the region indicated by the arrow has a
velocity of 75 m/s higher than the sur-
rounding tissue.
specimen (fig. 6a) shows similar characteristics
as discussed previously, i.e. uniform attenuation
and velocity over the field of view. In contrast,
the most distinctive feature of the infarcted tis-
sue (fig. 6b) is its overall microelastic in-
homogeneity. There are small regions of this
micrograph which show uniform velocity profiles
and attenuation characteristics (similar to normal
tissue), however, there also exists a number of
highly attenuating (darker) areas. These dark
areas show increases in sonic velocity of more
than 75 m/s compared to that seen in the neighbor-
ing tissue. The complex interference fringe pat-
tern indicates that, in the infarcted zones, there
is considerable variation in the sonic velocity
on a microscopic scale.
On several sections, ultrasonic attenuation
measurements were made at 100 MHz. The results
Table 1. Summary of 100 MHz attenuation
measurements.
Section Attenuation Number of
range (cm-i) measurements
A non-infarct 48 l
B non-farct^ 43b 2
B non-infarctc 41-68 2
C non-infarct 37-61 2
D infarct 38 - 100 5
E infarct 27-70 5
Off-axis component of sound is perpendicular
^to grain.
Off-axis component of sound is parallel to
^grain.
Typical value.
are presented in table 1. Sections A through C
were obtained from normal regions while sections
D and E were obtained from zones of infarction.
Each attenuation measurement was made on a small
region of tissue (1 x 10~'*cm2) gp^^ general,
several different areas on each section were
measured. The range of attenuation measured in a
section is given in column 2 of the table. The
variations tabulated here represent structural
variations in the tissue as opposed to uncertain-
ties in the measurement technique (which are on
the order of ± 10 percent). The number of mea-
surements made on each section was governed pri-
marily by the image contrast as seen on the TV
monitor. Areas that looked distinct on the moni-
tors were subsequently measured. Thus, sections
with uniform attenuation properties were subjected
to fewer measurements than sections displaying a
variety of brightness levels.
The attenuation values obtained in the 3 sec-
tions of normal tissue range from 37 to 68 cm"^ .
For section B, two entries are presented in the
table corresponding to the two different orienta-
tions of the section in the sound field. The
first entry, Bi, represents measurements made on
section B in the orientation depicted in figure
4a, while the second entry, B,,, presents measure-
ments made in the orientation depicted in figure
5a. The attenuation measured in the B orientation
show relatively minor variations from the typical
value of 43 cm--'. In contrast, measurements on the
same section in the B,, orientation yield values
ranging from 41 to 68 cm-^. Perhaps the most
striking feasure of the acoustic anisotropy is the
observation that the same area that yields an
attenuation coefficient of 45 cm-^ in orientation
Bi gives a value of 68 cm-^ in the B,, orientation.
Thus, variations in attenuation which are attrib-
uted solely to anisotropic propagation are of the
same magnitude as variations observed in different
sections .
The attenuation measured in the zones of in-
farction serve to quantify the visual impression
given by the acoustic micrographs. Infarcted
zones show inhomogeneous attenuation properties
with variations ranging between 27 and 100 cm"^
for the two sections analyzed. The highly attenu-
ating zones with attenuation coefficients in ex-
cess of 70 cm"^, are morphologically distinct from
any features present in the non-infarcted regions.
77
5. Discussion
This study presents some acoustic microstruc-
ture characteristics of myocardial tissue at a
frequency of 100 MHz. The features revealed in
the acoustic micrographs as well as the quantita-
tive attenuation and velocity data can lead to
important conclusions with regard to the complete
ultrasonic characterization of tissues.
First of all, 100 MHz acoustic micrographs can
be used to distinguish between and recognize in-
farcted and normal tissue. The most distinctive
feature of the infarcted zone is the high degree
of localized variations in attenuation and veloci-
ty. Such acoustic contrast could result from
fibrous scar tissue distributed in a matrix of
normal myocardium. The highly attenuating, high
sonic velocity regions can be attributed to areas
of substantial scar. This identification is con-
sistent with lower frequency measurements on tis-
sues rich in collagen [12].
Secondly, significant acoustic anisotropy,
arising from muscle fiber orientation, has been
observed at 100 MHz. One of the sections analyzed
(sample B) showed a large orientational dependence,
and this may have been due to the particular geom-
etry of the fibers. With a small component of the
propagation vector K along the fiber an increase
in attenuation was observed. This is in the same
direction as the attenuation anisotropy previously
observed in striated muscle at lower frequencies
[13]. The magnitude of change in attenuation is
rather substantial, 45 to 68 cm"^, indicating that
structural aspects and tissue architecture play an
important role in determining overall attenuation.
The orientation effect favors scattering type loss
mechani sm.
Thirdly, the large velocity variations measured
in the infarcted tissue may have important con-
sequences for low frequency attenuation measure-
ments. Phase cancellation losses have been im-
plicated as a primary source of artifact in at-
tenuation measurement made using piezoelectric
receivers [14,15]. Although small diameter
(~ .2 cm) receiving transducers or confocal pairs
of transducers can be used to minimize phase can-
cellation loss, they do not completely eliminate
it. Phase cancellation occurs when inhomogenei ties
in the tissue distort the ultrasonic phase fronts
presented to a spatially extended piezoelectric
receiving transducer. The i nterferogram, figure
6b, displays the high degree of wavefront distor-
tion which actually occurs in infarcted myocardium.
Quantitatively, the distortion is greater than 1.5
fringes or 540° in phase over a distance which is
smaller than the diameter of the small receiving
transducer (.2 cm) measurement reported at lower
frequency [2-4].
Lastly, it is interesting to compare the attenu-
ation measurements made at 100 MHz with those ob-
tained at lower frequencies. Figure 7 is a plot
of the attenuation coefficient divided by frequency
(a/f) vs f. The data in the frequency range 2 to
10 MHz were reported previously and are average
attenuation values obtained on fresh myocardium
excised 4 to 5 weeks after vascular occlusion [4].
On the other hand, the 100 MHz data, taken from
table 1, represent the range of attenuation values
in the infarcted and non-infarcted zones.
Analysis of figure 7 indicates that for normal
tissue, in the frequency interval 2 to 10 MHz, a/f
has a relatively constant value of .07 cm'^ MHz"i.
non-i nfarct
J ] I I \ L_
2 5 10 20 50 100
Frequency (MHz)
Fig. 7. Comparison of 100 MHz attenuation data
with that obtained at lower frequencies
for infarcted and non-infarcted myocardium.
Low frequency data are taken from the work
of Miller et al . [4] while the 100 MHz
data are from table 1.
The 100 MHz values reported here are substantially
higher (.37 to .68 cm'^ MHz'M- For the infarcted
tissue, the 2 to 10 MHz a/f values increase with
frequency. Linear extrapolation of these data
would predict a value of 1.9 cm~^ MHz"^ at 100 MHz.
However, the most highly attenuating regions ob-
served in the infarcted zones at 100 MHz have a/f
values of 1.0 cm"i MHz"^, a value significantly
lower than that predicted. Thus, the a/f versus
f dependence for normal and infarcted tissue pre-
dicted in the range 1 to 10 MHz may no longer hold
in the interval 10 to 100 MHz.
The following limitations to this conclusion
should be pointed out. First of all, the referred
to low frequency data were obtained on freshly ex-
cised tissue, while this 100 MHz investigation
used formalin fixed tissues. Although data ob-
tained on other tissues, e.g. , kidney, indicate
that the effect of formalin on the overall attenu-
ation values at 100 MHz may not be significant
[10], this has not yet been verified for the myo-
cardium. The second point concerns orientation
effects. The lower frequency measurements were
made by transmitting sound through the myocardial
wall roughly perpendicular to the epicardial sur-
face, while the 100 MHz measurements were done in
a plane perpendicular to this. Results presented
in this study indicate the presence of an orienta-
tional effect on the acoustic properties.
Although more comprehensive work needs to be
done to quantify the extent to which orientation
and fixing influence the acoustic properties, the
results of this study show that high frequency
acoustic microscopy can play an important role
in the understanding and characterization of tis-
sues at diagnostic frequencies.
Acknowledgment
The stimulating discussions with W. D. O'Brien,
University of Illinois are gratefully acknowledged.
References
[1] Lele, P. P. and Namery, J., A computer-based
ultrasonic system for detection and mapping
78
of myocardial infarcts, in Proc. San Diego [9]
Biomed. Symp. U, 121 (19721:
[2] Yuhas, D. E., Mimbs, J. W. , Miller, J. G.,
Weiss, A. N., and Sobel , B. E., Changes in
Ultrasonic Attenuation Indicative of Regional [10]
Myocardial Infarc'-'on, In Ultrasound in
Medicine , D. White, ed.. Vol . 3 (Plenurfi' Press ,
New York, 1977).
[11]
[3] Mimbs, J. W., Yuhas, D. E., Miller, J. G.,
Weiss, A. N., and Sobel, B. E., Detection of
myocardial infarction in vitro based on
altered attenuation of ultrasound. Circulation
Research (in press).
[4] Miller, J. G., Yuhas, D. E., Mimbs, J. W., [12]
Dierken, S. B., Busse, L. J., Weiss, A. N.,
and Sobel, B. E., Ultrasonic Tissue Charac-
terization: Correlation between Biochemical
and Ultrasonic Indices of Myocardial Injury,
in Proc. of IEEE Ultrasonic Symposium, J.
deKlerk and B. McAvoy, eds.. Cat. No. 76 [13]
Ch. 1120-4SU, p. 33 (Annapolis, 1976).
[5] Korpel , A., Kessler, L. W., and Palermo,
P. R., An acoustic microscope operating
at 100 MHz, Nature 232, 110 (1971).
[6] Kessler, L. W. , Korpel, A., and Palermo,
P. R. , Simultaneous acoustic and optical
microscopy of biological specimens.
Nature 239, 111 (1972). [14]
[7] Kessler, L. W. , Palermo, P. R. , and Korpel,
A., Practical High Resolution Acoustic
Microscopy, in Acoustical Holography, G. [15]
Wade, ed. , Vol. 4, p. 51 (Plenum Press,
New York, 1972).
[8] Kessler, L. W., Palermo, P. R., and Korpel,
A., Recent Developments with the Scanning
Laser Acoustic Microscope, in Acoustical
Holography, P. S. Green, ed.. Vol. 5, p. 15
(Plenum Press , New York, 1974).
Kessler, L. W., The Sonosmicroscope , in Proc.
of IEEE Ultrasonic Symposium, J. deKlerk,
ed., p. 735, Cat. No. 74-CH0-896-1 SU (IEEE,
New York, 1974).
Kessler, L. W. , VHP ultrasonic attenuation
in mammalian tissue, J. Acoust. See. Am. 53
1759 (1973).
Kessler, L. W., Tissue Characterization by
Means of Acoustic Microscopy, in Ultrasonic
Tissue Characterization, M. Linzer, ed..
Spec. Publ . 453, p. 261 (U.S. Government
Printing Office, Washington, D.C., 1976).
O'Brien, W. D., Jr., The Role of Collagen in
Determining Ultrasonic Propagation Proper-
ties in Tissue, in Acoustical Holography,
L. W. Kessler, ed. , Vol. 7 (Plenum Press,
New York, 1977).
Dussik, K. T. and Fritch, D. J., Determina-
tion of Sound Attenuation and Sound Velocity
in the Structures Constituting the Joints
and of the Ultrasonic Field Distribution
within the Joints on Living Tissues and
Anatomical Preparations, both in Normal and
Pathological Condition, Progress Report,
Project A454, Public Health Service, April
(1955), September (1956).
Marcus, P. N. and Carstensen, E. L., Problems
with absorption methods of inhomogeneous
solids, J. Acoust. Soc. Am. 58, 1334 (1975).
Busse, L. J., Miller, J. G., Yuhas, D. E.,
Mimbs, J. W., Weiss, A. N., and Sobel, B. E.,
Phase Cancellation Effects: A Source of At-
tenuation Artifact Eliminated by a CdS
Acoustoelectric Receiver, in Ultrasound in
Medicine, D. White, ed.. Vol. 3, pp. 1519-
1535"TPTenum Press, New York, 1977).
79
i 5
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed. , National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
ACOUSTIC PROPERTIES OF NORMAL AND ABNORMAL HUMAN BRAIN
F. W. Kremkau, C. P. McGraw, and R. W. Barnes
Bowman Gray School of Medicine
Winston-Salem, North Carolina 27103, U.S.A.
Attenuation and propagation speed at 1, 3, and 5 MHz and specific acoustic im-
pedance at 2 MHz were measured j_n vitro in 10 tissue samples from 8 abnormal human
brains. The results were compared to values resulting from a study of 5 normal
human brains reported elsewhere. Conclusions from literature review are that attenua-
tion is lower in gliobastoma than in normal brain and attenuation, propagation speed
and impedance are higher in meningioma than in normal brain. Conclusions from the
data of the present study are that subarachnoid hemorrhage appears acoustically
normal; infarcts have normal attenuation but increased speed; clots, intracerebral
hemorrhage, and metastases have increased attenuation and speed; hydrocephalic brain
has very low attenuation and low speed. In general, it appears that abnormal condi-
tions produce increases in all three acoustic properties studied.
Key words: Attenuation; brain tumor; clot; hemorrhage; hydrocephalus; impedance;
infarct; speed; ultrasonic.
1. Introduction
As part of a study of acoustic properties of
normal and abnormal tissues, attenuation and
propagation speed at 1, 3, and 5 MHz and specific
acoustic impedance (hereafter called impedance)
at 2 MHz were measured in vitro in 10 tissue
samples from 8 abnormal human brains.
Results from our study of 23 tissue samples
from 5 normal human brains are reported else-
where [l]i. The important observations were:
(a) attenuation measured with piezoelectric
transducers resulted in higher values
(0 to 50 percent) than those obtained
using radiation force method;
(b) attenuation (using transducer method)
was 0.9 dB/cm at 1 MHz and was a function
of f^"-' over the range 1 to 5 MHz;
(c) propagation speeds at 1 MHz were 1546
and 1539 m/s for fresh and fixed tissues,
respecti vely ;
(d) speed dispersion was 2 ms"^ MHz-^ and
1.6 ms"i MHz-i for fresh and fixed tissues,
respectively;
(e) white matter had attenuation 1.4 times
that for gray;
(f) attenuation of adult brain was 2.7 times
that for infant;
(g) one day aging reduced attenuation up to
20 percent;
(h) attenuation depended upon temperature to
the power -0.1 and -0.5 at 1 and 5 MHz,
respecti vely ;
(i) propagation speed as a function of tem-
perature exhibited a minimum at approxi-
mately 15 °C.
^Figures in brackets indicate literature
references at the end of this paper.
The differences between white and gray matter and
between infant and adult attenuation can be ex-
plained on the basis of tissue water content.
The minimum in propagation speed with respect to
temperature may indicate a sensitivity to the
crystal -1 iquid crystal phase transition for mem-
brane phospholipid bilayers.
There are several reports in the literature
regarding the acoustic properties of normal and
abnormal human brain tissue. A review of the
data for normal brain may be found in our earlier
report [1]. In addition, reported values for im-
pedance are 1.59 x 10^ MKS rayl (kg/m^s) for
"brain" [2], 1.60 for cerebellum [2], 1.51 for
cerebrum [3], and 1.52 for white matter [4].
An early paper by Wild and Reid [5] showed
differences in received echo patterns for normal
brain and several malignant and benign tumors.
More recently Fishman. Heyser, and Le Croissette
[6] reported high attenuation in tumor regions of
human brain sections using qualitative transmis-
sion images. A summary of the data for abnormal
brain is given in table 1. Only tumors are cited
as no other abnormality was found in the litera-
ture. Table 2 shows an attempt to simplify this
summary and draw some conclusions from it. It
lists reports of greater (+), equal (==), or lesser
(-) values for attenuation, speed, or impedance
as compared to the same authors' values for normal
brain tissue. Citing only those situations where
at least two reports are in agreement, the follow-
ing are concluded:
(a) attenuation is lower in glioblastoma than
in normal brain;
(b) attenuation is higher in meningioma than
in normal brain;
(c) propagation speed and impedance are
higher in meningioma than in normal brain.
81
Table 1. Attenuation, propagation speed, and impedance
for several brain tumors--l iterature summary.
Tissue Frequency Attenuation Speed Impedance Reference
(MHz) (dB/cm) (m/s) {^0'' MKS rayl )
2
7
2
7
8
9
9
7
2
7
9
7
10
11
12
13
2
9
9
2
7
3
9
9
10
12
13
14
3
12
2
7
10
4
7
7
7
3
2
apormalin fixed sampl es .
2. Materials and Methods
Normal and abnormal human brain tissue samples
were obtained at autopsy. Measurements were made
at 37 °C after at least 24 hours of fixation in
10 percent formalin. Except during measurement,
tissues were stored at 4 °C. Tissue samples were
cut to fill the sample holder - a Teflon® ring with
inner diameter 33 mm, outer diameter 45 mm, and
(sample) length 15 mm. Saran® membranes were
stretched over both sides of the chamber to pro-
vide flat tissue surfaces and acoustic windows.
The sample holder fit into a guide in a tempera-
ture controlled water bath which provided con-
sistent placement (within 1 mm in each coordinate)
relative to source and receiver transducers. The
transducers were one inch diameter Val pey-Fi sher
polished 1 MHz X-cut quartz. The source trans-
ducer was driven by an Arenberg PG-650C pulsed
oscillator. Bursts were typically 20 cycles long
and reasonably characterized by a single (funda-
mental) frequency. The receiver transducer was
connected to the input of a 7A18 amplifier in a
Tektronix 7904 oscilloscope with a 7B92 dual time
base. Amplitude of received burst, with and
without tissue sample between transducers, was
measured and attenuation coefficient, calculated
from the following relation:
Table 2. Attenuation, propagation speed, and impedance are higher
than (+), equal to (=), or lower than (-) that for normal
tissue as measured by the same authors. There is a con-
sensus that glioblastoma has lower attenuation and that
meningioma has higher attenuation, speed, and impedance.
Tumor Attenuation
Speed
Impedance Reference
acoustic neurinoma
+
+
2
7
arachnoidal sarcoma
2
astrocytoma
+
+
7
8
craniopharyngioma +
7
ependymoma
+
+
+
2
7
glioblastoma
+
7
10
11
12
13
g 1 i oma
2
medul loblastoma +
7
meningioma
+
+
+
+
+
+
+
+
+
2
7
3
10
12
1 3
14
metastatic adenocarcinoma
-
+
+
3
12
metastatic carcinoma
+
+
+
2
7
10
4
oligodendroglioma
7
pineal oma
7
pituitary adenoma
7
retinoblastoma
+
■
3
spongioblastoma
2
m = ^ 20 log^o
Ai "
where m and ni,^ are attenuation coefficients in
dB/cm of tissue sample and water, respectively,
m^ is window loss in dB, d is sample thickness in
cm, Aq is received amplitude with sample not
present (water path), and Aj is received amplitude
with sample present. Amplitudes were measured
with precision ± 2 percent or better and accuracy
(based on Tektronix specification) within 2 per-
cent. Measurements were taken at 1, 3, and 5
MHz. Corrections for m^ and m^ are normally less
than 1 percent except m^, at 3 and 5 MHz (typical-
ly 2 and 7 percent, respectively). The value for
m^- is obtained by measuring the received ampli-
tude change when a water filled chamber is in-
serted between the transducers. Additional re-
flection loss resulting from impedance discon-
tinuity between water and tissue is not signifi-
cant.
Propagation speed was calculated using the ar-
rival time change occurring when the sample was
inserted between source and receiver transducers,
c = d/(t2 - t ) where c is propagation speed in the
tissue sample^in m/s, d is sample thickness in
meters, tj is propagation time in seconds in d path
length of water, and t is arrival time change in
seconds when sample is introduced into sound path.
Arrival time changes are typically 200 to 400 ns
read to a precision of + 10 ns and accuracy (based
on Tektronix specification) within 5 percent.
Measurements were taken at 1, 3, and 5 MHz.
acoustic
neurinoma
arachnoidal
sarcoma
astrocytoma
craniopharyngioma
ependymoma
gl ioblastoma
6-8
gl ioma
medulloblastoma
men i ng ioma
4
5
2
2
5
1
1
5
metastatic
adenocarcinoma 1
metastatic 4
carcinoma 5
5
5
oligodendroglioma 5
pineal oma 5
pituitary adenoma 5
retinoblastoma
spongioblastoma 4
9-12
9
6-9
3.3a
20a
0.5
0.5
8-11
10-13
6.3a
0.9-1.2
0.6-1.2
35
0.3
8-10
4.3a
7
9
7
1522-1547 1
1530
1660
1545
1517a
1537-1545 1
1501
1525-1547 1
1529
1500
1540-1550 1
1640
1546-1569
1524a
1590
1535
1600
1532
57-1 .62
1.58
1.71
1 .64a
60-1 .62
1 .54
.57-1 .61
1 .56
1 .54
61-1 .64
1 .72
1 .57a
1.68
1 .60
1 .52
1 .67
1.59
82
Acoustic impedance was measured using the
method of Gregg and Palagallo [4]. A Parametrics
Pulser/Receiver 5050PR was used with a Hoffrel 310A
2.0 MHz, 13 mm diameter diagnostic transducer.
Display was produced on the Tektronix 7904 oscil-
loscope.
3. Results
The results of seven abnormal conditions are
given in table 3. With respect to attenuation
and propagation speed, the following observations
are made:
(a) subarachnoid hemorrhages appear normal
acoustically;
(b) infarcts have normal attenuation but
increased speed;
(c) clots, intracerebral hemorrhage, and
metastases have increased attenuation
and speed;
(d) hydrocephalic brain has very low at-
tenuation and low propagation speed.
Impedance values do not correlate well with tis-
sue type or with propagation speed values.
Table 3. Acourtic properties for seven abnormal conditions
in human brain.
Tissue
Attenuation
(dB/cm)
Speed
(m/s)
Impedance
(10'' MK5 rayl )
1 MHz
3 MHz
5 MHz
1 MHz
2 MHz
normals
0.9
2.8
5.2
1539
1 .43
metastatic lung
carcinoma
1.5
5.6
11.0
1568
1 .56
ventricular
clotb
2.3
9.6
12.2
1575
1 .55
subarachnoid
hemorrhage^
0.9
3.2
5.6
1535
1 .73
intracerebral
hemorrhage
1.9
8.2
12.8
1553
1 .78
hemorrhagic infarct
with necrosis
0.9
3.0
5.7
1552
1.72
Island of Kiel
infarct
0.9
3.2
6.6
1556
1 .46
hydrocephalus
0.2
0.8
1.3
1516
2.13
^See reference [1]. Impedance value was obtained by measurement
on one normal brain of the five studied with respect to attenu-
j^ation and propagation speed.
Mean of three samples.
Mean of two samples.
4. Discussion
In general, it appears that abnormal condi-
tions produce increases in all three acoustic
properties studied. The measured metastatic
carcinoma values agree reasonably well with
those of others [4,7] (see table 1). The low
attenuation for hydrocephalus is similar to
that reported for infant brain where high water
content was suggested as an explanation [1].
The low propagation speed for hydrocephalus was
not observed in the normal infant. Our results
show normal attenuation for infarcts (1 to 5
MHz). Mi 1 ler et al . [15] observed an increase
in attenuation for dog heart after infarct but
this was evidenced only at frequencies above
5 MHz. Lele and Namery [16] reported increased
attenuation above 2 MHz for infarcted dog heart
compared to normal. They also observed reduced
impedance for infarct which was not observed in
our study of brain.
Few of these abnormal tissues have been
studied by more than one group. Much more data
will be required before confident generaliza-
tions can be agreed upon.
Acknowledgments
The authors are grateful for technical assist-
ance from Ms. S. Gaskey.
This work was supported in part by NINCDS
Contract NOl-NS-4-2304.
References
[1] Kremkau, F. W., McGraw, C. P., and Barnes,
R. W. , Ultrasonic attenuation and preparation
speed in normal human brain, submitted for
publication, J. Acoust. Soc. Am.
[2] Schiefer, W. , Kazner, E. , and Kunze, S. ,
Clinical Echo-Encephalography , pp. 67-68
(Springer-Verlag, New York, 1968 ) .
[3] Ishikawa, S., Yukishita, K. , and Ito, K. ,
Ultrasonic attenuation in brain (7th report),
Jap. Med. Ultrasonics 3, 33 (1965), quoted
in reference 7.
[4] Gregg, E. C. and Palagallo, G. L., Acoustic
impedance of tissue. Investigative Radiol . 4,
357-363 (1969).
[5] Wild, J. J. and Reid, J. M. , The effects of
biological tissues on 15-mc pulsed ultra-
sound, J. Acoust. Soc. Am. 25^, 270-280 (1953).
[6] Fishman, L. S. , Heyser, R. C, and Le
Croissette, D. H., Ultrasonic transmission
measurements in human brain sections.
Radiology 112, 211-213 (1974).
[7] Ishikawa, S., Yukishita, K. , Sato, K. , Ito,
K. , and Wagai, T. , Ultrasonic attenuation in
brain tissue (the 7th report), Jap. Med.
Ultrasonics 3, 48 (1965).
[8] Uematsu, S. and Walker, A. E., A Manual of
Echoencephalography , p. 40 (Williams and
Wilkins, Baltimore, 1971).
[9] Van Venrooij, G. E. P. M., Measurement of
ultrasound velocity in tissue. Ultrasonics
9, 240-242 (1971).
[10] Tanaka, K. , Yukishita, K. , Ito, K. , Ehara,
K. , and Watanabe, H. , Ultrasonic diagnosis
of brain tumors, Proc. Intl. Symp. on Echo-
Encephalography, (1967), pp. 38-44, quoted
in Erikson, K. R. , Fry, J. J., and Jones,
J. P., Ultrasound in medicine - a review,
IEEE Trans. Sonics Ultrasonics SU-21, 144-
170 (1974).
[11] Le Croissette, D. H. and Heyser, R. D. ,
Attenuation and Velocity Measurements in
Tissue Using Time Delay Spectrometry, in
Ultrasonic tissue Characterization, M.
Linzer, ed.. National Bureau of Standards
Spec. Publ . 453, pp. 81-95 (U.S. Government
Printing Office, Washington, D.C., 1976).
83
[12] Oka, M. and Yosioko, K. , Ultrasonic absorp-
tion of human brain tissue. Paper #1302
presented at World Congress of Ultrasound
in Medicine, San Francisco (1976).
[13] Nakaima, N., Study of the ultrasonic at-
tenuation of brains bearing tumors, J^.
Wakayama Med. Soc. 27, 69-106 (1976).
[14] Kikuchi, Y., Tanaka, K. , and Wagai, T. ,
Early cancer diagnosis through ultrasonics,
J. Acoust. Soc. Am. 29, 824-833 (1957).
[15] Miller, J. G., Yuhas, D. E., Mimbs, J. W.,
Dierker, S. B., Busse, L. J., Laterra,
J. J., Weiss, A. N., and Sobel , B. E.,
Ultrasonic tissue characterization: cor-
relation between biochemical and ultra-
sonic indices of myocardial injury, IEEE
Ultrasonics Symposium Proceedings (1976)
IEEE Catalogue No. 76 CH 1120-5SU, pp. 33-43.
[16] Lele, P. P. and Namery, J., A Computer-Based
Ultrasonic System for the Detection and
Mapping of Myocardial Infarcts, in Proceed-
ings San Diego Biomedical Symposium, pp.
121-132, San Diego Biomedical Symposium,
San Diego, California, 1974.
84
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed., National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
FREQUENCY DEPENDENT ATTENUATION OF MALIGNANT BREAST TUMORS
STUDIED BY THE FAST FOURIER TRANSFORM TECHNIQUE
E. Kelly Fry, N. T. Sanghvi, and F. J. Fry
Indiana University School of Medicine and
Indianapolis Center for Advanced Research
Indianapolis, Indiana 46202, U.S.A.
and
H. S. Gallager
The University of Texas System/Cancer Center
Houston, Texas 77025, U.S.A.
The work discussed in this paper comprises one aspect of an experimental design
concerned with the use of multi-discipline examination methods to provide detailed
information on the interaction of normal and malignant breast tissues with a high
frequency sound field. The complete experimental design included x-ray and needle
biopsy examination of the breast of a patient prior to mastectomy, followed by: x-ray
examination of the excised breast (the malignant tumor remained intact in the excised
breast); ultrasound visualization and FFT studies of the formalin-fixed, excised
breast specimen; x-ray examination of 0.5 cm thick, whole breast sections of the tis-
sue; and, finally, sectioning and histological staining of the primary mal ignant
tumor region and other tissue areas of interest. Emphasis is given in this prelimi-
nary report to a study of attenuation of the sound beam as a function of frequency
for specific tissue paths (i.e., from skin to back surface of the excised breast),
which included (1) the malignant tumor, (2) the nipple and (3) the areola.
For the tissue path which included the malignant tumor, the FFT studies indicate
that the attenuation for the full range of frequencies studied (1.1 to 4.4 MHz) was
greater than that of any other area of the breast. A significant result of the in-
vestigations reported in this paper is the determination that this analytical tech-
nique is feasible and can yield data on malignant and normal regions of breast tis-
sue which correlate with the ultrasound visualization imaging information and with
the tissue structure information as revealed by histological examination.
Key words: Attenuation of areola; attenuation of breast tissue; attenuation of
malignant tumors; breast cancer detection; breast carcinoma; breast
examination techniques; FFT techniques for breast; histology of
breast tumors; signal processing for tissue; x-ray examination of
breast.
1. Introduction
At the present time, ultrasound visualization
techniques, if used in conjunction with other
clinical examination methods, can improve the
level of success in the differential diagnosis
of breast pathologies [1]^. However, the theo-
retical capability of ultrasound for providing a
successful method of early detection of breast
cancer is not presently realized, in part because
of the lack of sufficient clinical studies with
this method and, in addition, because of the need
for more basic experimental studies on the inter-
action of ultrasound and breast tissue. There
is, in fact, a serious need for relatively so-
^Figures in brackets indicate literature
references at the end of this paper.
phisticated investigations concerned with (1)
the complex structure of normal breast in terms
of its interaction with sound fields and (2) the
variability in structural features of malignant
tumors (including those in the same pathology
classification) insofar as this variability is
significant to medical diagnostic data obtained
by ultrasound examination of such tumors.
The experimental work discussed in this paper
comprises one aspect of an overall approach of
using multi-discipline examination methods, name-
ly, x-rays, ultrasound visualization, signal
processing of ultrasound transmission data (i.e.,
Fast Fourier Transform (FFT) techniques) and
histological techniques to provide detailed in-
formation on the structural features of breast
tissue and their interaction with a high frequency
sound field. In the initial experimental stages,
85
procedures such as the FFT are best performed on
an excised breast. A special preparation, namely,
an excised, formalin-fixed breast with a known in-
tact malignant tumor, was used in the investiga-
tions discussed in this paper. Detailed examina-
tion of such breasts, by both standard and ex-
perimental breast cancer detection methods, and
the correlation of the results with tissue struc-
ture information provided by histological in-
vestigations, can yield data that is pertinent
to early breast cancer detection [2-8]. The
rationale, validity and value of studying such
fixed, whole specimens by ultrasound visualization
techniques have been discussed in detail by Kelly
Fry and Gallager in a previous publication [8].
Additionally, Calderon et al . have published re-
sults of an investigation on the ultrasound at-
tenuation values of formalin-fixed breast tumors
[9].
In capsule form, the overall experimental de-
sign called for collection of data first by x-ray
examination of the breast of the patient prior to
surgery, followed by: x-ray examination of the
excised breast; detailed ultrasound visualization
and FFT studies of the formalin-fixed excised
breast specimen; whole breast sectioning of the
tissue in blocks of approximately 0.5 cm in thick-
ness; x-ray examination of each of these cross
sections and, finally, sectioning and histological
staining of the primary malignant tumor region and
other tissue areas of interest.
The FFT studies were considered in the
nature of a feasibility study and it is in that
regard they are reported in this paper. Emphasis
is given in this preliminary report to a study of
attenuation of the sound beam both in the region
of the malignant tumor and in other areas of the
breast tissue (such as nipple and areola) as a
function of frequency. As used in this paper,
the term "attenuation" designates all losses in
sound pressure amplitude or intensity as the
acoustic beam traverses the tissue, including
those due to specular reflection, scattering, re-
fraction, absorption and diffraction.
Both the location (as revealed by x-ray examina-
tion of the whole, excised breast) and the malig-
nant character of the primary mass (as revealed by
needle biopsy) were known at the time of the ultra-
sound visualization and FFT studies. At the time
of this writing, the histological classification
of the primary malignant tumor has been determined
and detailed histological studies are underway.
The precise correlations between information re-
vealed by histology, ultrasound visualization and
FFT techniques will be the subject of a later
paper.
2. Experimental Methods
The tissue specinen used in the subject study
was an excised, formalin-fixed, left breast of a
female subject who was 49 years of age at the time
of the mastectomy. Since the diagnosis of breast
carcinoma had been made by x-ray and needle biopsy
prior to surgery, the mastectomy was carried out
without a surgical biopsy so that the excised
breast contained the previously detected malignant
mass in an essentially undisturbed state. After
excision, the breast was made to assume its ap-
proximate normal contour by pinning it to a layer
of paraffin, was x-rayed and then formalin-fixed.
The fixation process consisted of covering the
pinned breast with 10 percent formalin solution,
completely draining this solution and adding a
fresh formalin solution every other day for a
total period of two weeks. After the completion
of this process, the breast was maintained in
formalin solution except for periods of ultrasonic
examination.
The instrumentation used for pulse-echo examina-
tion of the excised breast was a linear scan, B-
mode visualization system which included a variety
of types of transducers. Only two of these trans-
ducers were used as transmitters in the FFT
studies, namely, 1.1 MHz and 4.4 MHz center fre-
quency units. Most of the FFT studies were car-
ried out with the 4.4 MHz unit and the results ob-
tained with this transducer will be discussed in
this paper. This transducer has a diameter of
1.9 cm, a 7.5 cm focal length, a midband frequency
response of 4.4 MHz, a band width of 3 MHz jutt
above noise, and 1 MHz at the 3 dB point.
During the visualization scanning procedures,
the breast and examining transducers were com-
pletely immersed in mammalian Ringer's solution
at room temperature (23 °C) so that there was
direct fluid coupling of the ultrasound to the
tissue. Both transverse (medial-to-lateral) and
longitudinal (superior-to-inferior) scans were
taken at distance intervals of 1 to 2 mm. In
order to relate data recorded on the echograms to
specific areas of the tissue, an anatomical land-
mark on the tissue (such as the center of the
nipple or the center of a prominent skin discolor-
ation over the area of abnormality) was selected
and a highly reflective and attenuating acoustic
target placed on this landmark. With the focus
of the transducer set on the target center, the
echogram of the breast tissue showed a distinct,
easily identified front surface reflection and an
attenuation shadow of the target. Since the
linear coordinates (X and Y) were recorded for
each echogram, subsequent scans of the tissue with
the target removed could be directly related to
the chosen anatomical landmark. These surface
landmarks were subsequently used in the FFT ex-
periments for identification of desired sound
transmission tissue paths.
For the ultrasound attenuation studies in the
transmission mode, the previously described axi-
symmetric focused transducer was used as a trans-
mitter, and a 4 mm diameter, 10 MHz frequency,
PZT cera:mic sandwi ch-type piezoelectric probe was
used as a receiver. A Panametrics Model 5050R
unit pulse-excited the transmitter, and a series
unit step attenuator was used to attenuate the
ultrasonic received signals prior to their
entrance into an amplifier receiver. The breast
preparation was mounted on a three-motion co-
ordinate system in a temperature controlled mam-
malian Ringer's bath (average temperature 23 °C)
and positioned between the sending and receiving
transducers, with the anterior aspect of the tis-
sue facing the sending transducer (fig. 1). With
this arrangement, any area of the tissue could be
easily examined and the tissue could be moved in
1 mm steps in the three-axis coordinate system.
In order to record the system's reference wave-
form, transmitter and receiver transducers were
placed facing each other and adjusted in position
to receive the optimum plane wave acoustic signals.
A linearity test of the system was performed by
setting the attenuator unit at different values
and comparing the corresponding outputs of the FFT
86
Fig. 1. Experimental arrangement of examining
transducer (left), excised breast (side
view) and receiving transducer.
spectrum. The system was found to be accurate to
± 1 dB in the useable frequency band, for any
particular transmitting transducer. The received
ultrasonic waveforms were digitized at a sampling
rate of 10 or 20 nanoseconds using an eight bit
Biomation Model 8100 digitizer and then transfer-
red to a PDP-11/45 computer. Generally, four
waveforms were digitized for each tissue region
of interest (i.e., nipple, areola, etc.), stored
on a computer disc, and later processed using a
1024 point FFT. In order to obtain attenuation
and phase-angle versus frequency plots, they were
deconvolved with the original system's reference
waveform. The waveforms were transformed into the
frequency domain and the absolute amplitude for
each waveform was compared with the amplitude of
the reference waveform for the relative attenuation
measurement. To obtain the phase angle for each
waveform, the phase was compared with the refer-
ence waveform and subsequently linearized for each
frequency. For optimum frequency resolution, ap-
propriate sampling intervals were selected on the
Biomation digitizer. The output was presented in
a graphical form in three different formats; name-
ly, normalized pressure amplitude, attenuation and
phase-angle spectra. The phase-angle data is pre-
sented in a separate paper [10].
The tissue was sampled at various selected re-
gions by moving the specimen across the ultra-
sound beam on the three-axis system. The regions
of interest chosen for this study were specific
tissue paths (i.e., f"om skin to back surface of
the specimen) which included 1) the malignant
tumor, 2) the nipple and 3) the areola. In that
regard, eighteen separate transmission mode studies
were made in the tumor path, nine in the areola
path and eight in the nipple path. In addition,
the upper inner aspect of the breast was chosen as
representative of normal, middle-aged, mamnary
adipose tissue and test ultrasound transmissions
were made in that region.
Following completion of the FFT studies, the
tissue was longitudinally sliced to produce whole
breast sections from 5 to 8 mm in thickness. Each
of these cross sections was x-rayed and the radio-
graphs studied to determine which specific regions
of the breast would be further examined by means
of histological techniques. Tissue cubes 2 x 2.5
cm in overall dimensions were excised from the
selected regions of interest and prepared for
histological study. The primary malignant mass
was included in one of these cubes of tissue.
3. Results
The ongoing histological studies indicate that
the primary malignant mass was an invasive carci-
noma of duct cell origin with an intermediate de-
gree of differentiation. The mass was multinodu-
lar, fairly v/ell circumscribed but not encapsu-
lated, with a dense fibrotic center and a peri-
pheral shell (3 to 4 mm thickness) of neoplastic
cells. The fibrotic tissue, representing the
major tumor mass, was highly collagenous but the
cellular shell had a minimum deposit of collagen.
Included in the illustrations shown in this
paper are duplications of the recorded system
reference waveform (normalized pressure ampli-
tude versus frequency) for sound wave transmission
through Ringer's solution, compared to the wave-
forms recorded for sound transmission through
breast tissue immersed in Ringer's solution. Such
pressure amplitude graphical displays are present-
ed in order to demonstrate the characteristics of
the reference waveform and the dynamic range limi-
tations of the system.
Before discussing the results obtained for the
tissue path that included the malignant tumor, it
is of interest to consider the attenuation values
obtained for the areola and nipple tissue path re-
gions of the breast which, on the basis of the
x-ray examination, can be assumed normal. In
calculating precise attenuation values, on the
basis of the recorded, normalized pressure ampli-
tude values, the tissue path length was taken as
the distance between the anterior surface of the
skin overlying the nipple or areola and the poste-
rior surface of the breast specimen (as determined
by the previously obtained ultrasound visualiza-
tion images of the breast specimen). Therefore,
it is emphasized here that the attenuation values
shown in figures 3 and 4 are not specifically for
areola or nipple, but are for the tissue path that
includes these structures.
Figure 2 presents recorded pressure amplitude
waveforms for four sound transits through the
nipple at entrance points separated by 2 mm of
surface tissue. These waveforms show the maximum
variability in pressure amplitude (differences in
pressure amplitude for waveforms A, B, C, D) found
for the nipple region. The attenuation frequency
spectrum plot for the nipple path shown in figure
3 is based on the pressure amplitude waveforms
shown in figure 2 and the tissue path length (sur-
face of nipple to back surface of the specimen)
traversed by the sound wave.
Figure 4 is the attenuation frequency spectrum
for the areola region of the breast. As in the
case of the nipple, this data is also based on dif-
ferences in pressure amplitude response curves
(three sound transits through the areola region at
entrance points separated laterally by 1 mm of sur-
face tissue) and the tissue path length (areola
skin to back surface of the specimen). All other
sound transmissions in the areola path gave re-
sults within the range of values shown in figure
4. The tissue region of greatest attenuation, as
shown in figure 4, was located closer to the nipple
in comparison to the other two sound transit regions.
87
JL_ 1
12345678
Frequency Megahertz
Fig. 2. Computer recorded display of: (a) ref-
erence waveform for ultrasound trans-
mission through Ringer's solution only;
(b) waveforms A, B, C, D for ultrasound
transmission through four separate re-
gions of the "nipple tissue path" of an
excised, formalin-fixed breast immersed
in Ringer's solution.
9r
8 -
1 -
1 I I I I I
1 2 3 4 5 6
Frequency MHz
Fig. 3. Attenuation frequency spectrum
for "nipple tissue path" based
on waveforms shown in figure 2.
I I I I I I
1 2 3 4 5 6
Frequency MHz
Fig. 4. Attenuation frequency spectrum
for "areola tissue path" based
on three sound transmissions
through separate regions of the
areol a .
Frequency Megahertz
Fig. 5. Computer recorded display of: (a) ref-
erence waveform for transmission through
Ringer's solution only; (b) three wave-
forms for ultrasound transmission in
separate regions of the "tumor tissue path"
of an excised, formalin-fixed breast im-
mersed in Ringer's solution.
88
The wavefonn recordings shown in figure 5 re-
sulted from three sound transits in the tumor tis-
sue path, with each of the test points separated
by 1 mm of tumor tissue in terms of skin surface
distance. In contrast to results obtained in all
other tested regions of the breast, the recorded
waveform data show a total lack of recorded pres-
sure amplitude response in the higher frequency
regions. This result indicates that there is at
least 40 dB of amplitude loss, for the tissue path
from skin to back surface, in the frequency re-
gions above approximately 2 MHz. At the lower end
of the frequency spectrum, pressure amplitude re-
sponse was recorded but, as shown later in this
paper, the level of these pressure amplitudes in-
dicates a greater attenuation for the tumor tissue
path than for that of the nipple tissue path.
For the main body of the primary malignant
tumor, the overlying tissue is approximately
1.5 cm in depth (from skin to anterior border
of overt mass); the tumor has a diameter of
1.5 cm in depth and the tissue below the tumor
is approximately 2 cm in depth. X-ray examina-
tion of the excised breast indicated the tis-
sues surrounding the tumor were primarily fat-
ty in nature. Consequently, if the attenuation
for the 3.5 cm of tissue surrounding the tumor
is considered to be that of normal breast adi-
pose tissue and, further, if the attenuation
value for such tissue is assumed comparable to
that found in the present studies for tissue
paths (skin to back of tissue) through other
fatty regions of the breast, namely, 2.5 dB/cm
at 2.0 MHz, then the tissues surrounding the tumor
account for 8.8 dB of the attenuation. It should
be noted that the 2.5 dB/cm value for breast adi-
pose tissue is in general agreement with that
found by Calderon et al. [9], namely 2.3 dB/cm at
2.25 MHz for formalin-fixed, excised, normal
breast tissue. Deducting the 8.8 dB from the 40
dB maximum dynamic range response of the system
and taking into account the 1.5 cm tumor path, it
is found that the attenuation for the malignant
tumor itself is 21 dB/cm at a frequency of 2 MHz.
Since this value is in general agreement with that
of Calderon et a1 . who found an attenuation of
20 dB/cm at 2.25 MHz for formalin-fixed, excised,
malignant breast tumors, it is expected that the
40 dB dynamic range limitation, which represents
a factor of 100 in pressure amplitude and 10,000
in intensity, is close to an adequate dynamic
range response.
Carrying out the same type of calculation, using
the pressure amplitude data shown in figure 5 for
the lower frequencies, gives an attenuation value
for the malignant mass of 8 dB/cm at a frequency
of 1.5 MHz. This is calculated on the basis of
3.5 cm of normal fatty breast tissue, with an at-
tenuation value of 2.0 dB/cm at 1.5 MHz (as found
for other normal fatty regions of this breast
specimen) and a 1.5 cm depth of tumor mass.
Of the total of 18 separate runs through the
tumor tissue path, with the exception of two sound
transits in a particular region of this path, all
recorded results indicated that for frequencies
above 2 MHz there apparently was total attenuation
of the sound wave (i.e., within the 40 dB record-
ing level capability of the system). The excep-
tional cases showed 1) a pressure amplitude wave-
form that was highly, but not completely, attenuat-
ed by the tumor path and thus was within the 40 dB
maximum dynamic range of the system and 2) for a
sound transit in an area just 2 mm laterally dis-
tant from that of number one above, a pressure
amplitude waveform which had precisely the same
frequency spectrum, pressure amplitude response
as that found for normal fatty regions of the
breast. On the basis of this result, it was as-
sumed, prior to the histological examination, that
the tumor included a border region of less density
(conjectured to be a mixture of normal and neo-
plastic tissue) than the central mass and an ad-
jacent region of distinct adipose tissue. Based
on pressure amplitude data and the known tissue
path, an attenuation value of 13 dB/cm at a fre-
quency of 2 MHz was calculated for this border
area of neoplastic tissue.
The histological studies, to date, clearly show
adipose tissue immediately adjacent to and sur-
rounding the malignant mass, thus accounting for
the characteristics of the wavefonn which duplicat-
ed those of the fatty regions of the breast. The
3 to 4 mm thick neoplastic shell located adjacent
to the adipose tissue and surrounding the fibrotic
center of the tumor presumably accounted for the
13 dB/cm attenuation value.
4. Discussion
At the present time, in clinical studies con-
cerned with the use of pulse-echo methods for
breast examination, differential diagnosis is pri-
marily based on the characteristics of the wall of
the tumor, the presence and nature of the echoes
from the internal structure of the tumor and the
existence or absence of a shadowing phenomenon re-
sulting from the attenuation of acoustic energy
by a malignant mass. In evaluating each of these
parameters, considerable emphasis and reliance is
placed on the so-called "attenuation shadow" which,
when present, is generally considered to be indi-
cative of the presence of a malignant mass. How-
ever, precise knowledge regarding attenuation
characteristics of normal, benign and malignant
breast tissue is extremely limited. This is a
serious deficit since its consequence may be mis-
diagnosis of the benign or malignant nature of a
breast tumor.
Considerable care should be used in interpret-
ing attenuation shadows produced by pulse-echo
examination of breast, with specific regard to the
question of whether normal or benign tissues can
produce significant shadowing effects. In that
regard, Kobayashi has found that fat necrosis re-
sults in an attenuation shadowing comparable to
that produced by a malignant mass, as judged by
pulse-echo visualization [11]. Calderon et al . ' s
studies at a frequency of 2.25 MHz of formal in-
fixed excised breast tissue gave values of 20 dB/
cm for malignant masses, 9 dB/cm for benign masses
and approximately 2.3 dB/cm for normal breast tis-
sue surrounding the tumors [9]. Although there
is a significant difference between attenuation
values for benign and malignant tumors as found by
these investigators, it must be realized that in
the clinical pulse-echo methods, differential diag-
nosis is made on the basis of a comparison between
the image pattern of the overt mass and that of
the surrounding normal tissue. Therefore, the
Calderon et al . data indicate that for pulse-echo
techniques in which the sound beam makes two pass-
es through the breast, common sized benign tumors
(2 cm and above) with an attenuation of 9 dB/cm
could result in shadowing in relation to surround-
89
ing normal tissue with an attenuation value of
2.3 dB/cm. Additionally, the first author of this
paper recently found, in the case of several clini-
cal patients, each with an overt, benign mass
within the breast (identified by x-ray examination
and confirmed by surgical biopsy), that the echo-
grams obtained using standard pulse-echo methods
at frequencies of 4.4 MHz to 5.0 MHz displayed
distinct shadows in the region of the mass. The
significance of this result, insofar as the present
paper is concerned, is the recognition of the need
for quantitative attenuation data on multiple
types of benign and malignant breast tumors, as
opposed to simple visual inspection of shadow
phenomena of such masses and consider. -^ion of
whether signal processing techniques may orovide
some of the needed data [12].
The FFT studies discussed in this paper, for
the tissue path which included the malignant tumor,
showed that the attenuation for the full range of
frequencies studied (1.1 to 4.4 MHz) was greater
for this tissue path than for that of any other
area of the breast. Further, these findings were
consistent with those found in the visualization
echograms, that is to say, even at a frequency of
1.1 MHz, this specific tumor could be recognized
by "pale shadowing," while at the higher frequen-
cies the attenuation shadow was more opaque and
formed a distinctive border in relation to ad-
jacent tissue.
From the viewpoint of possible in vivo breast
examinations by FFT methods, it is of interest
that in the present study the dual structural com-
ponents of the tumor (as shown by the two attenua-
tion values) and the adipose tissue surrounding
the tumor were recognized prior to any histologi-
cal investigations. Further, the finding that
the tc' ^1 malignant mass was attenuating but the
greater attenuation was in the region of fibrosis
agrees with earlier studies [8] relating attenua-
tion (as judged by nonquantitati ve pulse-echo
methods) of breast tumors with specific tumor tis-
sues (as revealed by detailed histological
studies). The highly collagenous nature of the
primary malignant mass combined with its high at-
tenuation values and the minimum collagen content
of the surrounding cellular shell is of interest
in regard to the relative importance of elastic
properties versus physical structural make-up for
sound reflection and transmission through overt
tissue masses [13]. The ongoing histological
studies and the preliminary phase-angle data may
provide some information on this aspect [10].
There are many factors, both tissue and instru-
mentation related, associated with the attenuation
in the nipple path region as determined by the FFT
ultrasound transmission studies discussed in this
paper. Included in the tissue aspects are the non-
uniform structure of the skin surface and the vari-
able tissue components within the nipple and in
the regions deep to the nipple. Although the
above factors may be significant, information
which is relevant to present pulse-echo clinical
techniques can be derived from the preliminary
FFT data without complete analysis of such param-
eters. In that regard, for example, the results
shown in figure 3 are important in relation to
detection of an overt breast mass located in the
region directly posterior to the nipple. If pulse-
echo techniques are applied to scan directly over
the nipple region, there may be a lack of success
in detecting such masses (particularly at frequen-
cies of the order of 5 MHz) because of their loca-
tion in the attenuating path of the nipple. For
the purposes oftumor detection, therefore, the
tissues deep to the nipple should be viewed by
scanning from the side regions of the breast and
by using lower frequencies. These two approaches
to visualizing structures deep to the nipple were
successfully applied by the present authors in the
case of the excised breast discussed in this paper,
and breasts studied in vivo.
To accomplish the overall aim of the studies
carried out in this preliminary investigation,
that is, to correlate x-ray, visualization, FFT
analysis and histological findings, requires ex-
tensive and time-consuming investigations. How-
ever, if some of the significant parameters that
are relevant to differential diagnosis of breast
tumors by ultrasonic techniques are to be deter-
mined by methods other than the necessai^'iy slow
process of interpreting ultrasound clinical data,
such detailed investigations are necessary. A
significant result of the investigation reported
in this paper is the determination that this com-
puter based, signal processing technique is feas-
ible and can yield data on malignant and normal
regions of breast tissue. It also appears evident
from these experiments that similar investigations
could be carried out on freshly excised whole
breast tissue. If such investigations are suc-
cessful, then this technique could be considered
for application to breast patient examination.
Success in that regard could lead directly to de-
velopment of an ultrasound breast examination
technique which might eventually limit the need
for surgical biopsy to a small number of excep-
tional cases.
Acknowledgments
Grateful acknowledgment is made to the assist-
ance of George W. Ga ler, ultrasound technologist,
in all aspects of the above studies.
This research was supported by the Grace M.
and Ralph W. Showalter Residua»"y Trust, and by
the Indianapolis Center for Advanced Research, Inc.
References
[1] Kobayashi, T., Review: Ultrasonic Diagnosis
of Breast Cancer, in Ultrasound in Medicine
and Biology, Vol. 1 (1975), pp. 383-391.
[2] Gallager, H. S. and Martin, J. E., The Patholo-
gy of Early Breast Cancer, in ' ~east Cancer:
Early and Late, The Univers: f Texas,
M. D. Anderson Hospital, pp. 3/-50 (Year Book
Medical Publishers, Inc., Chicago, 1968).
[3] Gallager, H. S. and Martin, J. M., The study
of mammary carcinoma by mammography and whole
organ sectioning. Cancer 23 (4), 855-873
(1969).
[4] Fry, E. Kelly, Franklin, T. D., Jr., and
Gallager, H. S., Ultrasound Visualization of
Excised Breast Tissue: An Experimental Ap-
proach to the Problem of Precise Identifica-
tion of Structure from Echogram Data, Acous-
tical Society of America meeting, Washington,
D.C., April (1971).
[5] Fry, E. Kelly, Gallager, H. S., and Franklin,
90
T. D. , Jr. , In Vivo and In Vitro Studies of
Application o7 Ultrasonir~Visualization Tech-
niques for Detection of Breast Cancer, in
Proc, IEEE Ultrasonics Symposium, Miami,
Florida, December (1971). ~
[6] Martin, J. E. and Gallager, H. S., Reflec-
tions on Benign Disease: A Radiographic-
Histologic Correlation, in Early Breast
Cancer Detection and Treatment, H. S. Gal-
lager, ed. , pp. 177-181 (John Wiley and Sons,
New York, 1976).
[7] Fry, E. Kelly, The Use of Ultrasound Methods
to Detect Changes in Breast Tissue Which Pre-
cede the Formation of a Malignant Tumor, in
Acoustical Holography, L. W. Kessler, ed.,
pp. 1-20, Vol. 7 (Plenum Publishing Corp.,
New York, 1977).
[8] Fry, E. Kelly and Gallager, H. S., A Research
Approach to Visualization of Breast Tumors by
Ultrasound Methods, in Ultrasound: Its Ap-
plication in Medicine and Biology, F. J. Fry,
ed. (Elsevier Scientific Publishing Co. ,
Amsterdam, 1978).
[9] Calderon, C, Vilkomerson, D., Mezrich, R.,
Etzold, K. F., Kingsley, B., and Haskin, M. ,
Differences in the attenuation of ultrasound
by normal, benign and malignant breast tis-
sue, J. Clin. Ultrasound 4 (4), 249-254
(19761:
[10] Fry, E. Kelly, Sanghvi , N. T., Fry, F. J.,
Gardner, G. W., and Gallager, H. S., Deter-
mination of Alterations of Phase Angle of
Ultrasound Transmitted Through a Malignant
Breast Tumor: A Preliminary Investigation,
in Ultrasound in Medicine, D. White, ed..
Vol. 4 (Plenum Publ ishing Corp., New York,
1978).
[11] Kobayashi, T., Takatani, 0., Hattori, N.,
and Kimura, K., Study of sensitivity-graded
ultrasonotomography of breast tumor (pre-
liminary report), Med. Ultrasonics 10 (1),
38-40 (1972).
[12] Lizzi, F., Katz, L., St. Louis, L., and
Jackson-Coleman, D., Applications of spectral
analysis in medical ultrasonography. Ultra-
sonics 14 (2), 77-80, March (1976).
[13] Fields, S. and Dunn, F., Correlation of
echographic visual izabil ity of tissue with
biological composition and physiological
state, J. Acoust. Soc. Amer. 54 (3), 809-
812 (1973T:
91
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed., National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
CORRELATION OF ULTRASONIC ATTENUATION WITH CONNECTIVE TISSUE
CONTENT IN BREAST CANCERS
Toshiji Kobayashi
Division of Clinical Electrophysiology
Department of Internal Medicine
National Cancer Center Hospital
5-1-1, Tsukiji, Chuo-ku, Tokyo-104, Japan
In recent years, ultrasonic techniques have been applied to the diagnosis of breast
cancer. Various characteristics suggestive of tumor pathology have been determined.
In this paper, some of these signs, such as the acoustic middle shadow and the com-
plete disappearance of the distal limit of the tumor mass echo, are correlated with
the connective tissue content of various breast cancers. Because of the high acoustic
impedance of connective tissue relative to the other tissues, tumors rich in connec-
tive tissue are believed more strongly to attenuate the ultrasonic waves and thus to
cause shadows in the tissues beyond them.
Key words: Attenuation; breast cancer; cancer; connective tissue; differential
diagnosis; medullary carcinoma; papillary carcinoma; scirrhous
carcinoma; shadowing; ultrasound.
1. Introduction
Because of its noninvasive nature, diagnostic
ultrasound is a good clinical tool for the fine
visualization of soft tissue pathologies. This
technique has been applied in the past several
years to the differential diagnosis of breast
tumors with high diagnostic accuracies. Echo-
graphic criteria for differentiating malignant
and benign lesions have been reported by several
investigators [1-8]^.
Echographic patterns of shadowing beyond tumors
(retromammary shadowing) provide reliable and ac-
curate diagnostic information for the diagnosis of
breast cancer. In this paper, various malignant
signs, such as the acoustic middle shadow sign and
the complete disappearance of the distal limit of
the tumor mass echo, are shown to be correlated
with the connective tissue content within the
breast cancer. Because of the high acoustic im-
pedance of connective tissue relative to the other
tissue, tumors rich in connective tissue are be-
lieved more strongly to attenuate the ultrasonic
waves and thus to cause shadows in the tissues be-
yond them.
2. Method and Results
During the past five years, 1617 cases of pal-
pable breast tumors were examined echographi cal ly
and all cases with breast cancer were verified by
mastectomy. This material forms the basis of this
retrospective study. The method of examination is
^Figures in brackets indicate literature
references at the end of this paper.
shown in figure 1. The criteria for the differen-
tial diagnosis of breast tumors can be divided in-
to three major categories comprising (1) the bound-
ary echoes and shape, (2) internal echoes, and
(3) shadowing, as illustrated in figure 2.
Fig. 1. Method of scanning the breast. The 5 MHz
probe is moved automatically along the arc
of a circle, and ultrasonic coupling to the
patient is through the water in the flexible
plastic bag.
93
BOUNDARY ECHO & SHAPE
usual ly regular and
smooth, round, oval
or hetnioval
INTERNAL ECHO
RETROMAMMARY SHADOWING
uniform- si zed,
homogeneous or
echo-free (anechoicl
irregular and jagged
bizarre, crab-like or
polymorphous
non-uniform-sized ,
heterogeneous or
polymerous
tadpole-tail sign
lateral shadow sign
accentuation of posterior echo
acoustic middle shadow
(posterior shadowing)
attenuation of posterior echo
Fig. 2. Diagram illustrating echogram appearances for differential diagnosis
of benign and malignant tumors.
The first two categories, that is, the boundary
echoes and shape, and internal echoes, result from
echo patterns mainly confined to the tumor mass it-
self or its vicinity. The third category, shadow-
ing, is a result of various attenuation mechanisms
such as multiple reflections, changes in beam velo-
city, and scattering due to the impedance discon-
tinuities within the tumor.
The boundary echoes provide important diagnostic
information, producing outlines which are usually
irregular in the case of breast cancers, and regu-
lar in the case of benign lesions, such as cysts
and fibroadenomas. A cancerous tumor is uneven
with a jagged edge, sometimes triangular, rectangu-
lar or irregular in shape, whereas a benign tumor
is usually round, oval or hemioval. Moreover, the
internal echo pattern in a malignant tumor is
usually irregular both in size and distribution, in
contrast with the either homogeneous or anechoic
internal echo pattern of a benign lesion.
Other differential diagnostic criteria result
from echo patterns due to multiple reflections from
the tumor mass, modifying the shadowing. It is
this phenomenon which is the principal subject of
this paper. The tadpole-tail sign arises from the
low energy loss in the beam as it passes normally
through the distal wall of a cyst leaving suffi-
cient energy to cause multiple reflections or "ring-
ing" between the distal cyst wall and the chest
wall. The difference in the velocities in tissue
and cyst may play an important role in causing this
pattern. This sign is usually not seen in the
majority of malignant lesions except in some cases
of medullary carcinoma. The lateral shadow sign is
believed to be formed by the almost total reflec-
tion of the ultrasonic beam at the lateral wall of
cystic lesions as, for example, with benign tumors
[6]; an explanation of the formation of this sign
has recently been published [9].
A characteristic of malignant lesions designated
as "acoustic middle shadow sign" is sometimes ob-
served. This phenomenon is believed to be caused
by the high attenuation of the ultrasonic energy by
malignant tumor tissue, due to large acoustic im-
pedance discontinuities within such tissue.
The reliability of these various signs for the
differential diagnosis of various histological types
of breast cancer is seen in table 1. Diagnostic
accuracy is greatest in scirrhous carcinoma and
lowest in medullary carcinoma.
Echographic characteristics suggestive of malig-
nant lesions, such as the acoustic middle shadow
sign and the complete disappearance of the distal
limit of the tumor mass echo, have been correlated
with the connective tissue content of tumor masses
for various histological types of breast cancer.
The grading of the connective tissue content was di-
vided into three categories based on the observation
of many microscopic field specimens. These cate-
gories were expressed as follows: 1) the connective
tissue over 75 percent (rich), 2) 25 to 75 percent
(moderate) and 3) less than 25 percent (poor). The
specimens were stained by hemtoxylin and eosin and
are shown magnified by approximately 100.
The results of the analysis are shown in tables
1, 2, 3 and 4. The acoustic middle shadow sign as
well as the complete disappearance of the distal
limit of the tumor mass echo, were most frequently
seen in the cases with abundant connective tissue,
whereas the tadpole-tail sign and the lateral
shadow sign were often seen in cases with little
connective tissue content, mainly in medullary
carcinoma. These analyses were carried out on 12
cases of various breast cancers (4 scirrhous car-
cinomas, 5 papillary carcinomas and 3 medullary
carcinomas).
94
Table 1. Reliability of various criteria for the differential diagnosis
of breast cancer (53 cases).
4-> O
ro -r-
s- -c
ra
01 S-
Q- O
Q- E
=C 3
Type of carcinoma
Medullary
Papi 1 1 ary
Sci rrhous
Diagnostic accuracy rate
20/24
(83°/,)
13/15
(87%)
14/14
(100%)
nant
Complete disappearance of
distal limit of tumor echo
17/24
(70%)
12/15
(80%)
11/14
f -t no/ \
(78%)
Irregular boundary echo
23/24
(95%)
14/15
(93%)
11/14
(78%)
ID
s:
Acoustic middle shadow sign
18/24
(75%)
11/15
(73%)
14/14
(100%)
c
Bilateral disappearance of
distal limit of tumor echo
4/24
(17%)
1/15
(0.6%)
0/14
(0%)
(U
Tadpole-tail sign
4/24
(17%)
2/15
(1.3%)
0/14
(0%)
CO
Lateral shadow sign
4/24
(1.3%)
2/15
(1.3%)
0/14
(0%)
Table 2. Connective tissue content of 12 cases of carcinoma. Rich = over 75 percent;
moderate = 25 to 75 percent; poor = less than 25 percent. High consistency is
indicated by (+++), intermediate consistency by (++), and low consistency by (+).
Connective tissue content
Nonconnecti ve tissue content
1)
Scirrhous
carcinoma
(Tl)
(+++)
2)
Sci rrhous
carcinoma
(Tl)
(+++)
3)
Scirrhous
carcinoma
(T2)
(+++)
4)
Scirrhous
carcinoma
(T2)
5)
Papi 1 1 ary
carcinoma
(Tl)
6)
Papillary
carcinoma
(Tl)
7)
Papi 1 1 ary
carcinoma
(T2)
8)
Papi 1 lary
carcinoma
(T2)
(+++)
9)
Papi 1 1 ary
carcinoma
(T2)
(+++)
10)
Medul lary
carcinoma
(Tl)
(+)
11)
Medul lary
carcinoma
(T2)
(+)
12)
Medullary
carcinoma
(T2)
(+)
rich
moderate
(++)
(++)
(++)
(++)
poor
(+++)
(+++)
(+++)
rich
moderate
(++)
(++)
(++)
(++)
poor
(+)
(+)
(+)
(+)
(+)
Table 3. Correlation between appearance of various echo signs and connective tissue contents in
12 cases of carcinoma, classified according to echo signs. Signs indicated by circled
letters in the table have the higher reliabilities. A = acoustic middle shadow sign;
C = complete disappearance of distal limit of tumor mass echo; N= no change of shadow
or intermediate pattern; T = tadpole-tail sign; L = lateral shadow sign.
Connective tissue content
rich moderate poor
1)
Scirrhous
care i noma
(Tl)
(A)
N
T
L
A
C
N
T
L
A
C
N
T
L
2)
Scirrhous
carcinoma
(Tl)
@
N
T
L
A
C
N
T
L
A
C
N
T
L
3)
Scirrhous
carcinoma
(T2)
C
N
T
L
A
(0
N
T
L
A
C
N
T
L
4)
Sci rrhous
carcinoma
(T2)
A
c
N
T
L
A
c
N
'X)
CD
A
C
N
T
L
5)
Papillary
carcinoma
(Tl)
A
c
N
T
L
A
c
N
(T)
A
C
N
T
L
6)
Papi 1 lary
carcinoma
(Tl)
A
c
N
T
L
A
c
N
(T)
®
A
C
N
T
L
7)
Papi 1 lary
carcinoma
(T2)
A
c
N
T
L
A
c
IN)
T
L
A
C
N
T
L
8)
Papil lary
carcinoma
(T2)
(A)
©
N
T
L
A
c
N
T
L
A
C
N
T
L
9)
Papil lary
carcinoma
(T2)
CA)
©
N
T
L
A
c
N
T
L
A
C
N
T
L
10)
Medullary
carcinoma
(Tl)
A
c
N
T
L
A
c
N
T
L
A
C
N
8
(T)
11)
Medul 1 ary
carcinoma
(T2)
A
c
N
T
L
A
c
N
T
L
A
C
N
0
12)
Medullary
carcinoma
(T2)
A
c
N
T
L
A
c
(N)
T
L
A
C
N
T
L
95
Table 4. Correlation between appearance of various echo signs and connective tissue
contents in 12 cases of carcinoma, classified according to carcinoma type.
(S) = scirrhous carcinoma; (P) = papillary carcinoma; (M) = medullary
carcinoma.
Connective tissue content Nonconnecti ve tissue content
rich
moderate
poor
rich
moderate
poor
Acoustic middle shadow sign
(S)
(P)
(S) (S)
(P)
(S)
(P)
(S) (S)
(P)
Complete disappearance of dis-
tal limit of tumor mass echo
(S)
(P)
(S) (P)
(S)
(S)
(S)
(P)
(S) (P)
Tadpole-tail sign
(P) (P)
(M) (M)
(M) (M)
(P) (P)
Lateral shadow sign
(P) (P)
(M) (M)
(M) (M)
(P) (P)
No shadow or intermediate
pattern
(P) (M)
(P) (M)
3. Demonstration of Typical Echograms
and Tissue Characteristics
Tables 3 and 4 show the correlations between
the amount of connective tissue and the various
echo patterns described in the following para-
graphs .
A. Scirrhous Carcinoma (Tl and T2)
A series of echograms recorded by the sensitivity-
graded method [6] show typical patterns of shadow-
ing (acoustic middle shadow sign) suggestive of T2
malignancy (fig. 3).
The distal limit of the tumor mass echo gradual-
ly fades as the attenuation increases, as may be
Fig. 3. Echograms of breast with T2 scirrhous
carcinoma (acoustic middle shadow sign)
recorded at a range of sensitivities, to-
gether with corresponding histological
section magnified approximately 100 times.
seen on the echograms taken at -15 to -20 dB at-
tenuation, and finally completely disappears. The
microscopic picture shows the tissue type content
to be predominantly connective in cellular struc-
ture. The echogram in figure 4 is a typical pat-
tern of early scirrhous carcinoma (Tl), showing the
acoustic middle shadow underneath the tumor mass
echo. Histologically, the connective tissue com-
ponent is predominant.
Fig. 4. Echogram of breast with Tl scirrhous
carcinoma (acoustic middle shadow sign),
together with corresponding histological
section magnified approximately 100 times.
96
Fig. 5. Echogram of breast suggesting Tl scirrhous
carcinoma (acoustic middle shadow sign),
together with confirmatory corresponding
histological section magnified approxi-
mately 100 times.
The echogram in figure 5 shows another typical
acoustic middle shadow pattern suggestive of Tl
malignancy. The microscopic picture also shows
rich connective tissue.
These results suggest that rich connective tissue
content may play an important role in producing the
acoustic middle shadow sign and in the complete
disappearance of the distal limit of the tumor mass
echo in clinical echograms of scirrhous carcinoma.
B. Papillary Carcinoma (Tl)
The echogram of early papillary carcinoma shows
a typical attenuation of the shadowing, so that the
acoustic middle shadow sign is not clear. More-
over, the typical benign signs such as tadpole-tail
sign and the lateral shadow sign do not appear
either (fig. 6. ) .
The histological picture shows connective tissue
and nonconnecti ve tissue approximately evenly dis-
tributed. Therefore, this pattern of papillary
carcinoma may lie in between those of scirrhous
carcinoma and medullary carcinoma (see C below)
with respect to the connective tissue content.
This kind of evenly-distributed connective and non-
connective tissue produces an equivocal echographic
pattern which makes it difficult correctly to diag-
nose early papillary carcinomas.
C. Medullary Carcinoma (T2)
This echogram shows a rather homogeneous in-
Fig. 6. Echogram of breast with papillary carci-
noma (note absence of acoustic middle
shadow and tadpole-tail signs), together
with corresponding histological section
magnified approximately 100 times.
ternal echo, mimicking a benign cyst, and an atypi-
cal tadpole-tail shorter than that of tadpole-tail
usually seen in the case of benign tumors (fig. 7).
Consequently correct diagnosis is rather difficult.
The histological picture shows rich non-connective
tissue content, and especially abundant and uni-
formly distributed tumor cells. This kind of
homogeneous and even distribution of tumor cells
should produce little sonic reflection from within
the tumor mass, thus mimicking the pattern of
benign cysts.
4. Discussion
In the past several years, researchers have
hypothesized that the ultrasonic characteristics
of malignant breast tumors could be attributed to
the increased attenuation of ultrasonic energy by
the cancerous tissue with greater internal acoustic
impedance discontinuities, and hence increased
scattering losses, as compared with those with
normal breast tissue and benign breast tissues.
In direct measurements of the attenuation of
ultrasound within 18 samples of various normal,
benign and malignant breast tissues, significant
differences have been found [10]. At 2.25 MHz,
malignant tissue had the highest attenuation (1.2
dB per wavelength), whilst benign tissues had a
median value of 0.6 dB per wavelength. These re-
sults support the hypothesis.
From a histological standpoint, the percentage
of connective tissue (consisting of fibroblasts
97
Fig. 7. Echogram of breast with T2 medullary
carcinoma (the tadpole-tail sign is
shorter than that generally seen with
benign tumors), together with correspond-
ing histological section magnified ap-
proximately 100 times.
and fibrous tissue) has been suggested to be one
of the principal factors responsible for ultrasonic
attenuation in neoplastic tissue [11].
In an analysis [12] of 53 cases of breast cancer
for the incidence of the appearance of various echo-
graphic characteristics such as the acoustic middle
shadow sign, the complete disappearance of the
distal limit of the tumor mass echo was suggestive
of malignancy, and the tadpole-tail sign and the
lateral shadow sign was suggestive of a benign con-
dition. Diagnostic accuracy rates were 100 per-
cent, 87 percent and 83 percent in scirrhous carci-
noma, papillary carcinoma and medullary carcinoma
respectively, as may be seen in table 1. It is in-
teresting to note that the tadpole-tail sign and
the lateral shadow sign appeared most rarely, and
the disappearance of the distal limit of the tumor
echo was most frequent, in medullary carcinomas.
Generally these signs suggest benign lesions, and
this may account for the fact that the lowest diag-
nostic accuracy is obtained with medullary carci-
nomas .
The connective tissue content seems to be gener-
ally rich in scirrhous carcinoma and poor in medul-
lary carcinoma. The tumor cellular component
usually predominates Within the nonconnecti ve tis-
sue of medullary carcinoma and this tissue struc-
ture may be responsible for the bioacoustical homo-
geneity observed within the tumor mass. This homo-
geneity in medullary carcinoma may produce less at-
tenuation of ultrasonic energy, resulting in shadow-
ing suggestive of benign tumors, such as the
tadpole-tail sign and the lateral shadow sign. In
the present investigation, these echographic charac-
teristics appear regardless of the tumor size and
the location of the tumor mass within the breast.
Medullary carcinoma has lower attenuation than
other types of carcinoma [10]; its measured value
at 2.25 MHz falls in a range comparable to that of
benign tumor tissue, whereas the attenuation is
highest in scirrhous carcinomas followed by papil-
lary carcinomas. It is interesting to note that
the discrimination on the basis of attenuation was
very clear between normal breast tissue, benign
tumors, and malignant tumors, except for the case
of medullary carcinoma. This finding is in agree-
ment with the hypothesized echographic origin of
the medullary carcinomas even at the frequency of
5 MHz used in the present analysis.
As to the other factors contributing to the
echographic characteristics of tumors, especially
irregular and nonsmooth boundary echoes of various
carcinomas, the incident rate of its irregularity
is 95 percent for medullary carcinoma, 93 percent
for papillary carcinoma and 78 percent for scirrhous
carcinoma. This kind of irregularity may be close-
ly related with infiltrative pattern of cancerous
cells, and the reflection coefficient at the tumor
interface as well as other factors.
It should be emphasized that the correlation of
clinical echographic signs with the histological
characteristics of the tumors does not imply that
in any given case, a particular sign is indicative
of any histological type. Because of inadequacies
of the clinical equipment used in this study, these
correlations should only be considered qualitative.
The principal conclusion is that the amount of con-
nective tissue within the tumor mass may play an
important role in the formation of the acoustic
middle shadow sign and the complete disappearance
of the distal limit of the tumor mass echo, and
that it is probably due to the increased ultrasonic
attenuation caused by the high acoustic impedance
of the connective tissue.
References
[1] Baum, G., Ultrasonic Examination of the
Breast, in Fundamentals of Medical Ultra-
sonography, G. Baum, ed., pp. 380-402
(Putnam' s Sons, New iork, 1975).
[2] Cole-Beuglet, C. and Beique, R. A., Con-
tinuous ultrasound B-scanning of palpable
breast masses. Radiology 117, 123-128 (1975).
[3] Fujii, T. , Izuo, M., Kishi, S., Yokomori , T.,
and Fujimori, M., The results and the evalua-
tion of ultrasonic diagnosis of breast dis-
ease, J. Jap. Soc. Cancer Therap. 8^, 253-
[4] Hirose, M. and Furuki, R. , Ultrasonic diag-
nosis of breast disease, Proc. 17th Meeting
Jap. Soc. Ultrasonics Med. 17, 37-38 (1971)
[5] Jellins, J., Kossoff, G., Reeve, T. S., and
Barraclough, B. H., Ultrasonic grey scale
visualization of breast disease. Ultrasound
Med. & Biol. I, 393-404 (1975).
Kobayashi , T. , Takatani, 0., Hattori , N.,
and Kimura, K. , Differential diagnosis of
breast tumors: The sensitivity graded method
of ul trasonotomography and clinical evalua-
tion of its diagnostic accuracy. Cancer 33,
940-951 (1974).
Pluygers, E., Diagnostic ultrasonore, par
echographie A et B, des affections Mammaires,
J. Beige. Radiol. 58, 15-29 (1975).
Wagai, T., Tsutsumi , M. , and Takeuchi, H. ,
Diagnostic Ultrasound in Breast Diseases, in
Present and Future of Diagnostic Ultrasound,
I. Donald and S. Levi, eds., pp. 148-161
Okujima, M., Refraction of ultrasonic beam
incident near circumference of spherical
medium, Proc. Jap. Soc. Ultrasonics Med. 29,
231-232 (1976).
[10] Calderon, C. , Vilkomerson, D., Mezrich, R. ,
Etzold, K. E., Kingsley, B., and Haskin, M.,
Differences in the attenuation of ultrasound
by normal, benign and malignant breast tissue,
J. Clin. Ultrasound 4, 249-254 (1976).
[11] Field, S. and Dunn, F., Correlation of echo-
graphic visualization of tissue with bio-
logical composition and physiological state,
J. Acoust. Soc. Am. 54, 809-812 (1973).
[12] Kobayashi, T. , Takatani, 0., Hattori, N. ,
and Kimura, K. , Clinical investigation for
the differential diagnosis of breast tumor
by means of the sensitivity graded method
of ul trasonotomography (fourth report) -
Analytical re-appraisal of diagnostic
criteria in echographie and mammographic
findings, Proc. 24th Meeting Jap. Soc.
Ultrasonics Med. 24, 147-148 (1973).
99
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed., National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
THE ATTENUATION OF SELECTED SOFT TISSUE AS A FUNCTION OF FREQUENCY
D. H. Le Croissette, R. C. Heyser, P. M. Gammell , and J. A. Roseboro
Jet Propulsion Laboratory
California Institute of Technology
Pasadena, California 91103, U.S.A.
and
R. L. Wilsoni
School of Medicine
University of Southern California
Los Angeles, California 95033, U.S.A.
Measurements of attenuation versus frequency have been made on pancreas, kidney,
fat and liver specimens of hog tissue. Hogs were of the "Berkshire" variety and the
tissue obtained from a slaughterhouse. The measurements were taken under controlled
conditions of temperature and formalin fixing to determine the effects of aging and
fixing on the tissue. A swept-f requency transmission system was used operating over
the range of about 1.5 to 9.5 MHz. The curves show the changes occurring in fresh
tissue as a result of several days of aging at refrigerator temperatures (5 °C) and
the effects of fixing the tissue in formalin immediately post-mortem or after several
days of 5 °C storage. All measurements were made at 37 °C.
Key words: Attenuation; tissue properties; transmission; ultrasound.
1. Introduction
The recent interest in characterizing soft tis-
sue by ultrasound is based upon measurements made
as early as 1952 [2]^ and later observations by
scientific investigators attempting to lay the
groundwork for a new technique of ultrasonic diag-
nosis. The work reported here is the start of a
systematic attempt to measure the attenuation of
soft tissue as a function of frequency. This
topic was chosen because of the belief, substan-
tiated by initial measurements [2], that there is
I a variation in the attenuation of ultrasound
through tissue as a function of frequency that can
be correlated with the pathological state of the
tissue. The measurements reported here were made
j on excised tissue obtained from hogs. This paper
I reports data obtained from fresh, aged and fixed
specimens.
In a typical pulse-echo ultrasound imaging
system, the frequency variation of attenuation
cannot usually be identified. Furthermore, the
ii increase in attenuation of normal tissue as a
j function of frequency tends to mask the high
1 frequency behavior of tissue when a measurement
is made over a long path length, i.e., far from
the transducer. These two factors have made the
identification of pathological tissue by its
^Current address: Harbor General Hospital,
Torrance, California 90509.
^Figures in brackets indicate literature
references at the end of this paper.
frequency-dependent characteristic unlikely and
so this phenomenon plays little part in the sub-
jective pattern recognition techniques used in
contemporary diagnostic ultrasonography. It is
believed, however, that if an identification
mechanism based on frequency-dependent phenomenon
can be found, suitable instrumentation will be
devised for tissue identification.
Measurements on tissues of known pathologies
can greatly increase the knowledge of tissue
characterization needed to develop such instru-
mentation. Specimens obtained from cadavers
are especially useful since the individual organ
can be studied. Most autopsy material, however,
is not available until after a one or two-day
delay. Furthermore, the general availability of
pathological tissue is greatly enhanced if a
study is not constrained to specimens a few hours
old. This study is therefore to determine the
acoustic attenuation characteristics of several
organs as a function of time after death and
after fixing in formalin solution.
2. Methodology
There are two major methods for examining the
ultrasonic propagation behavior of tissue as a
function of frequency. The most popular method
[3-5] uses the frequency spectrum that is emitted
by a heavily damped transducer operated in a
pulse mode. This method is capable of giving
useable data at frequencies several times the
natural resonant frequency of the crystal with
adequate signal-to-noise ratios. By contrast.
101
the technique used in this work does not pulse
the transducer but sweeps the operating frequency
over a wide band. This technique [6], known as
Time Delay Spectrometry (TDS), reduces the ef-
fects of multipath propagation and reverberation
by using a swept frequency transmitter and re-
ceiver so that the frequency acts as a "time
tag" to the signal. The sweep of the receiver
is delayed from that of the transmitter so it
will only respond to the signals with the proper
time delay. This discriminates against signals
with different arrival times, especially those
due to reverberation and multipath propagation.
The data were taken with a standard swept-type
spectrum analyzer-tracking generator which has
been modified by offsetting the tracking generator
by a constant frequency. A Panoramic SPA-3 spec-
trum analyzer and Panoramic G6 tracking generator
were used. This tracking generator can be adjust-
ed to compensate for the time of transit of the
ultrasound through the sample. A sweep rate of
100 MHz'S"-' and 3 dB intermediate frequency band-
width of 600 Hz were used. If the sweep frequency
versus time of the spectrum analyzer is linear,
this constant frequency offset will exactly com-
pensate for a constant time delay of the ultra-
sound through the specimen. The time delay will
be constant across the frequency sweep if both
the specimen and the medium in which it is im-
mersed are nondispersi ve. Dispersion of the me-
dium and the sample will not affect the measure-
ments by more than 1 dB, provided the average
velocity of the sample and medium does not change
by more than 1 percent. Linearity requirements
on the sweep of the spectrum analyzer are ex-
treme, however.
The measurements were made over the frequency
range 1.5 to 9.4 MHz. Two Aerotech Alpha trans-
ducers were used. The 10 MHz, 0.63 cm diameter
(0.25 inches) transducers were selected so that
the resonant frequency was just above the range
of interest. This frequency was chosen so that
the transducer sensitivity increased with fre-
quency over the band. This compensates for the
increase in tissue attenuation with frequency to
give a more nearly constant signal-to-noise ratio
over the selected frequency spectrum. The speci-
men was located midway between the transducers,
which were 12 cm apart.
3. Specimen Handl ing
Specimens of tissue from hogs were studied no
more than four hours after slaughter. The tissue
type and the time sequence of the measurements
are shown in figure 1. All specimens were heat
sealed in a plastic bag containing either 0.9
percent saline or formaldehyde diluted 10:1 with
water (i.e., approximately 4 percent). The liver
specimens were sliced to approximately 15 mm
thickness; the kidney and pancreas specimens were
studied intact. The backfat specimen was left in
the form of a slab as it was stripped from the
carcass. The thickness of all of the specimens
was calipered at the locations studied. Fat is
normally removed one day after slaughter which
accounts for the one-day delay shown in figure
Kb).
Figures 2 through 9 show the measured attenua-
tion versus frequency for the hog tissue at
37 °C. Four or five locations on each specimen
were studied. Obvious ducts, vessels, and taper-
DAYS POST MORTEM
SPEC. NO.
0 1 1 1 2 1 3 1 4 1 5 1 6 1 7 1 8 1 9 1 10 1 11 1 12 1
1 1 1 — . —I 1 . — L — - I 1 I 1 1 1 1 I
PANCREAS ^
KIDNEY
r.
I 205 9
O 1 1 1 1 1 0 1 , . , .1 , , i, lot 1 1 ;Sl
1 2062
1 1 1 1 1 O ) lit 1 1 t .^-^
1 2061
1 1 O 1 1 1 t 1 ) 1 1 ) 1
1 2060
' ■ . l ■ i ■ 1 t . . . , .V-,', , . . ^. .V-. J^. -\-, ,V A ^K\^\\\^
1 2047
O 1 1 O 1 1 1 0 1, . f :\X7. } t
] 2064
o 1 1 o 1 1 1 0 f 1 lot I t t ..^
1 2048
1 1 O 1 } t 1 1 J 1 I 1 i .. aN
1 2049
i A 1 t t i t lot 1 1 t §1
1 2063
1 1 1 1 1 t t t O t t 1 1 NX\1
1 2065
11 1 1 |o| 1 oi,,, t,„J„^,,t
(a)
h^AX'xv^NXKx^'^l FIXED (:':-:^::v:-:';-:';''->v:'J FROZEN
1 1 5 °C Q DATA
DAYS POST MORTEM
SPEC. NO.
0|l|2|3l4l5|6|7|8|9llO|n|l2|
FAT -
1 2055
lot 1 o 1 1 1 oi-i?*:' 1 lot 1 1
1 2057
1 1 1 1 1 1 o'iixi' t 1 1 1 1
1 2056
1 1 1 c ! 1 1 ! t ! 1 1 1 1
1 4.:<i;.<.t.,.J,.-4.,,i<v.J,, J A 1
LI m
1 2050
1 o 1 lo! 1 1 o^<^:V^^^^^^^^,;^t*^^^^^^^^^^ ,1
1 206 7
o 1 1 o 1 1 Id 1 t o 1 1 1 1 ,';5l
1 2052
1 1 1 1 1 o 1 1 1 1 1 1 i
I 2051
1 1 o'l r ! 1 1 i t 1 1 1'^^
1 2Q66
1 \t ' 1 r t 1 1 i c 1 1 i 15^
1 2053
1 1 1 1 1 ! 1 1 o 1 1 i, , t_:^
1 2054
1 t t 1 1 f 1 1 o 1 i !
1 2068
1 1 i 1 1 i 1 O i 1 1 1 t OS)
(b)
Fig
. 1.
t" J FIXED [ : 1 FROZEN
1 1 5 °C O DATA
Specimen history: (a) pancreas and
kidney, (b) fat and liver.
ed edges were avoided. In the case of the kidney,
measurements were not made in the central portion,
which contains the large vessels and the renal
pelvis. All of the kidney measurements were made
in the region of the pyramids, and so include
both cortical and medullary matter. The loca-
tions studied on any given specimen can be iden-
tified by the shape of the symbol used on figures
2 through 5. That is, the circle, square, or
triangle represents the same visible location on
the specimen when studied on the day of slaughter,
five days later, and after fixing. Care was
taken to avoid multipath effects by visual in-
spection of the area of the specimen and by
observing the spectrum analyzer for obvious signs
of multipath interference. Since no fine struc-
ture was observed in the attenuation versus fre-
quency curves, the data are reported at 0.5 MHz
i nterval s .
Each specimen was measured at four or five
points. This is a compromise between sampling
enough locations to insure that the measure-
ments are representative and reducing experi-
ment time to a minimum. Since tissue is known
to deteriorate rapidly at 37 °C, it is impor-
tant that the experiment time be kept as short
as possible. Stray air bubbles that initially
clung to the specimen were gently brushed away.
No gas evolution was subsequently observed
vi sual ly.
102
2 4 6
Frequency, MHz
10
25r
20-
5 15-
10-
4 6
Frequency, MHz
10
25r
20-
15-
10-
25
20
15
53
u lOh
M-
1 Oh -
20-
I— 0,
2 4 6 8
Frequency, MHz
(d)
A A
□
□
o
O V
a <s
□ O
□
Fig. 2.
Attenuation versus frequency for hog pancreas: (a) 6 hours
post-mortem (fresh) , (c) 5 hours post-mortem (fixed), (d) 5
4 6 8
Frequency, MHz
post-mortem (fresh), (b) 5 days
days post-mortem (fixed).
10
These measured values of tissue attenuation
exhibit considerable variability. At this
stage in the experiments it was considered that
little was to be gained by applying statistical
analysis on the four or five readings taken al-
though it is recognized that a mean and vari-
ance will be meaningful on readings from a
) higher number of points. Accordingly, the
; averaged or summarized data are presented in
I the form of band diagrams in figures 6 through
J 9. These bands were selected to follow the data
with no more than 10 percent of the points fall-
i ing outside of the band.
4. Data Acquisition
The response of the system was recorded with
attenuations of 0, 5, 10, 15, 20, 25, and 30 dB
switched into the system. The response with the
specimen between the transducers was then record-
ed on the same polaroid film. Since the sweep
of the spectrum analyzer only covers a 5.5 MHz
range in the present equipment, it was necessary
to take the data in two parts. As a check,
readings taken from both spectra are reported in
the range of overlap, which is 4.5 to 5.5 MHz.
If only one reading is reported at a given loca-
103
OL- 0,
4 6
Frequency, MHz
2 4 6
Frequency, MHz
25
20
S 15
10
25
20
15
10
OL
2 4 6
Frequency, MHz
10
Frequency, MHz
Fig. 3.
Attenuation versus frequency for hog kidney: (a) 3 hours post-mortem (fresh),
post-mortem (fresh) , (c) 4 hours post-mortem (fixed), (d) 5 days post-mortem (f
(b) 5 days
i xed ) .
tion on the specimen then the readings from the
upper and from the lower spectra agree within
0.5 dB. Since the calibration spectra are ob-
tained for 5 dB increments, interpolation to the
nearest 1 dB is reliable. For smoothness in
plotting the data, the graphs are plotted to the
nearest 0.5 dB, although no significance should
be attached to changes of less than 1 dB. The
accuracy of the data is believed to be generally
± 1 dB, with rare deviations of ± 2 dB. Meas-
urements using standard, non-biological samples
show a system repeatability error of + 0.5 dB.
5. Measurements
Figure 2 shows the plot of attenuation versus
frequency for hog pancreas under four conditions;
(a) six hours post-mortem (fresh), (b) five days
post-mortem (fresh), (c) fixed five hours post-
mortem, and (d) fixed five days post-mortem.
Storage of the fresh tissue at 5 °C results in a
decreased attenuation particularly at the higher
frequencies. These data do not seem conclusive,
however, since figure 2(b) shows a bimodal dis-
tribution of readings. The organ lacks rigidity
and so accurate placement and replacement of the
104
25r
20-
15-
10-
25r-
25|-
20-
15-
15-
."10-
- o -
10-
o) 5 "
OL- OL
2 4 6
Frequency, MHz
4 6
Frequency, MHz
8
V o
A □ i o
r\ o
O g 8 °
□ □ O
25
20-
15
10
2 4 6
Frequency, MHz
10
OL- 0,
2 4 6
Frequency, MHz
10
Fig. 4.
Attenuation versus frequency for hog backfat: (a) 1 day post-mortem (fresh), (b) 6 days
post-mortem (fresh) , (c) 1 day post-mortem (fixed, (d) 6 days post-mortem (fixed).
transducers on these specimens is very difficult.
In addition, the pancreas contains lobules of
glandular tissue on the order of 1 cm, inter-
spersed with fat. It is therefore considered
that the wide spread in data can be accounted for
by the lack of rigidity and the inhomogeneity of
the specimen. Figure 6 shows the bands of atten-
uation versus frequency for the fresh specimens
under two conditions; six hours and five days
post-mortem.
The measurements on fresh and fixed kidneys
are shown in figure 3 and the corresponding bands
of attenuation are given in figure 7. For the
fresh specimens, storage of the tissue at 5 °C
for five days has resulted in a decrease in at-
tenuation, especially at the higher frequencies.
Measurements taken 48 hours post-mortem (not re-
ported here) lie between the two values. This
decrease is just sufficient to separate the band
curves. The attenuation of fixed tissue is
above that of the tissue just post-mortem.
Figures 4 and 8 made on hog fat show that there
is little effect of aging on the attenuation char-
acteristics. Little significance can be placed on
the differences at low frequencies shown in figure
8 since the attenuation is only of the order of
105
25
25
15-
sio-
(a)
4 6
Frequency, MHz
20
15
10
10
15
10
(b)
X
^ a
C) □
2 4
Frequency,
6
MHz
10
251-
20
15
10
OL-
Frequency, MHz
4 6
Frequency, MHz
Fig. 5.
Attenuation versus frequency for hog liver: (a) 5 hours post-mortem (fresh), (b) 5 days
post-mortem (fresh) , (c) 5 hours post-mortem (fixed), (d) 5 days post-mortem (fixed).
the uncertainties in the measurements. Fixing the
specimens gives a slightly higher attenuation.
There is considerable difference between this fat
and typical fat found in human studies so that
these data should be applied with caution. Hog
fat is white and of a uniform texture whereas
human fat is yellow and has a lobular structure
with connective tissue interspersed.
Figures 5 and 9 show the data for hog liver.
Here the storage of a fresh specimen results in
a lower attenuation over a five-day period. The
fixed specimens show increased attenuation. The
data bands are fairly narrow probably because of
the homogeneity of the tissue at the resolution
level studied.
These tissue specimens were stored at 5 °C and
heated to the measurement temperature of 37 °C
over about a one-hour period. Following this
procedure, each specimen was returned to the re-
frigerator for storage until the next measurement
was made or until the specimen was fixed. For
comparison, some specimens were held undisturbed
at 5 °C until a single measurement was taken. No
significant difference in attenuation could be
seen between the two groups.
106
T
6 h post-mortem
5 d post-mortem
4 6
Frequency, MHz
10
Attenuation versus frequency data bands
for hog pancreas: 6 hours and 5 days
post-mortem (fresh).
5
3 h post-mortem
iyZ\ 5 d post-mortem
4 6
Frequency, MHz
10
Fig. 7,
Attenuation versus frequency data bands
for hog kidney: 3 hours and 5 days
post-mortem (fresh).
T
Frequency, MHz
8. Attenuation versus frequency data bands
for hog backfat: 1 day and 6 days
post-mortem (fresh).
T
5 h post-mortem
5 d post-mortem
Frequency, MHz
Fig. 9. Attenuation versus frequency data bands
for hog liver: 5 hours and 5 days
post-mortem (fresh).
107
6. Discussion
The accuracy of these measurements relies
upon the internal sweep linearity of the spectrum
analyzer to provide a constant time delay across
the frequency band matching the constant transit
time of the sound energy through the specimen.
Slight non-linearities in the sweep were observed
to give an error as high as 1 dB in the measured
attenuation. Some indication of this error may
be seen in the region between 4.5 and 5.5 MHz.
In practice, two sets of data were taken; the
lower half extended from 1.5 to 5.5 MHz and the
upper part from 4.5 to 9.5 MHz. Small errors in
the overlapping region are indicated by two val-
ues being plotted with the same symbol. In many
cases identical values were obtained or the er-
ror was small. Since these measurements were
taken, a new system has been built in which the
sweeps of the transmitter and receiver are ob-
tained from digitally programmed phase-coherent
synthesizers. This will insure that the system
is aligned over the entire frequency range of 1
to 10 MHz and will materially improve the quality
of the data .
Acknowledgment
This paper presents the results of one phase
of research conducted at the Jet Propulsion
Laboratory, California Institute of Technology
for the National Science Foundation, by agree-
ment with the National Aeronautics and Space
Administration.
References
[2] Le Croissette, D. H. and Heyser, R. C,
Attenuation and Velocity Measurements in
Tissue Using Time Delay Spectrometry, in
Ultrasonic Tissue Characterization, M.Linzer,
ed.. National Bureau of Standards Spec. Publ.
453, pp. 167-196 (U.S. Government Printing
Office, Washington, D.C., 1976).
[3] Lele, P. P., Mansfield, A. B., Murphy, A. I.,
Namery, J., and Senapati , N., Tissue Charac-
terization by Ultrasonic Frequency-Dependent
Attentuation and Scattering, in Ul trasoni c
Tissue Characterization, M. Linzer, ed..
National Bureau of Standards Spec. Publ. 453,
pp. 167-196 (U. S. Government Printing
Office, Washington, D.C., 1976).
[4] Lizzi, F. L. and Laviola, M. A., Tissue
signature characterization using frequency
domain analysis. Ultrasonics Symposium
Proceedings (IEEE), pp. 714-719, 1976.
[5] Miller, J. G., Yuhas, D. E., Mimbs, J. W. ,
Dierker, S. B., Buse, L. J., Laterra, J. J.,
Weiss, A. N. , and Sobel , B. E., Ultrasonic
tissue characterization: correlation be-
tween biochemical and ultrasonic indices of
myocardial injury. Ultrasonics Symposium
Proceedings (IEEE), pp. 33-43, 1976.
[6] Heyser, R. C. and Le Croissette, D. H.,
A new ultrasonic imaging system using Time
Delay Spectrometry, Ultrasound in Med, and
Biol . 1, 119-131 (1974).
[1] Wild, J. J. and Reid, J. M., Further pilot
echographic studies on the histologic struc-
ture of tumor of the living and intact human
breast. Am. J. Pathol. 28, 831-861 (1952).
108
CHAPTER 4
SCATTERING AND ATTENUATION
109
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed.. National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
CLINICAL SPECTRUM ANALYSIS TECHNIQUES FOR TISSUE CHARACTERIZATION
Frederic L. Lizzi and Marek E. Elbaum
Riverside Research Institute
80 West End Avenue
New York, New York 10023, U.S.A.
A spectrum analysis system is being used in a clinical measurement program to ex-
amine the reflectance characteristics of intraocular structures. The on-line system
computes power spectra descriptive of echoes returned from well-defined tissue seg-
ments chosen by the examiner. Clinical data are digitized for subsequent post-
processing and cataloguing. This paper presents an overview of the theoretical
analysis used to formulate system design and to guide data interpretation. The
analysis accounts for stochastic tissue architectures and treats the specific il-
lumination conditions employed in the system. The paper also discusses system param-
eters as related to observable tissue characteristics and presents typical test
data and clinical spectra for several pathologic conditions.
Keywords: Clinical ultrasound; ocular tumor; power spectra; Rayleigh scattering;
ultrasonic spectrum analysis.
1. Introduction
Within the past few years, a variety of ad-
vanced techniques have been employed to measure
the ultrasonic properties of soft-tissue struc-
tures [l-3]i. This paper describes a clinical
spectrum analysis technique [4] which has been
used to measure the frequency-domain charac-
teristics of ultrasonic backscatter from tissues
of the eye and orbit. The technique employs a
clinical, on-line system together with computer
post-processing and data cataloguing. Following
laboratory verification, an extensive series of
clinical measurements has been carried out. The
in vivo spectral data base is now being evalu-
ated in terms of potential diagnostic signifi-
cance and is also being employed in analytic model-
ling of tissue reflectance.
The spectrum analysis techniques have been de-
signed to accommodate the stochastic nature of
clinically relevant tissue structures, such as
ocular tumors and vitreous hemorrhages. These
structures exhibit randomness in terms of the size,
orientation, and/or spatial position of their con-
stituent internal scattering centers. Single re-
flectance measurements made on such stochastic
entities exhibit a significant degree of statisti-
cal fluctuation which can preclude the reliable
identification of pertinent tissue characteristics.
Accordingly, the spectrum analysis techniques de-
scribed below incorporate on-line ensemble averag-
ing: the resulting power spectra determinations
constitute statistically stable descriptors of tis-
sue properties. In addition, compensatory proce-
dures are employed to account for extraneous spec-
tral weighting introduced by the transfer functions
^Figures in brackets indicate literature
references at the end of this paper.
of transducers and electronic subsystems.
The spectral techniques can ultimately be em-
ployed clinically to examine organs other than the
eye. However, their in vivo application is now
practicable in ocular examinations since the media
of the eye do not present additional constraints
found in other organs. If transmission through
the ocular lens is avoided [5], acoustic attenua-
tion and refraction do not present significant
obstacles in viewing intra-ocular structures at
center frequencies higher than 10 MHz. The large
bandwidths that can be realized at these high fre-
quencies permit spectral sensitivity to morpho-
logical features whose scale sizes are significant-
ly less than 0.5 mm. In addition, the narrow beam-
widths (e.g. , 0.3 mm) available at these center
frequencies provide precise spatial definition of
tissue segments to be analyzed and permit meaning-
ful ensemble averaging, even within small tissue
vol umes .
To date well over 200 clinically derived spectra
of normal and abnormal structures have been ac-
quired, processed, and catalogued in disease-
indexed digital files. Previous publications have
discussed the operation of the spectrum analysis
system and presented clinical findings [6,7].
This paper first outlines the analytic framework
which has guided system design, clinical operation,
and data interpretation. It then describes system
operation emphasizing the relations between system
performance parameters and observable tissue charac-
teristics. Lastly, it presents representative test
results for comparison with clinical data obtained
for several ocular pathologies.
2. Outline of Analytic Framework
The clinical deployment of the spectrum analysis
techniques employed in the present system has been
111
guided by an analysis which accounts for both the
stochastic nature of tissue architectures and the
geometry and temporal spectrum of the ultrasonic
illumination beam. This analysis is presented in
outline form to facilitate discussions in follow-
ing sections. (A more comprehensive treatment will
be presented elsewhere [8].)
The analysis is discussed, first, in terms of
its general formulation which accounts for the use
of focused illumination conditions. Second, tis-
sue architectures which are statistically station-
ary are discussed. Lastly, attention is given to
tissues which can be modelled as randomly distribut-
ed Rayleigh scatterers.
A. Basic Formulation
The geometry involved in the analysis is shown
in figure 1. A focused transducer emits a short,
broadband ultrasonic pressure pulse. Radio fre-
quency echo signals received from scatterers within
a range-gated segment (0-L) are subjected to spec-
trum analysis. The range gate is positioned within
the focal volume of the transducer and its time
duration is chosen to be much larger than the dura-
tion of echoes from a single scatterer. The trans-
mission characteristics of all intervening struc-
tures are assumed to be statistically homogeneous
and to result in negligible attenuation.
X
Gated
Segment
Fig. 1. Geometry for spectrum analysis
measurements.
widths (e.g. , 0.3 degree) and small fractional
range segments, L/R, that are of interest, the
beam pattern over the gated segment can be expres-
sed as a function of cross-sectional arguments
only: i.e., F(re) = F(y,z). Also, the range, r,
can be replaced by R + x. Thus, eq. (1) can be
rewritten as
„-jwR/c ■ ,
p'(co) = (jcoA/c)P(co)^ e J"''/''F(y,z) . (2)
2ttR
The backscatter from the gated tissue volume is
characterized in terms of a scattering function,
S(a); x,y,z). This function relates the amplitude
of the incident pressure at a point (x,y,z) within
the volume of interest to the strength of the back-
scattered spherical wave originating at that point.
The backscattered wave is sensed by the transducer
which responds to the spatial average of the pres-
sure distribution generated over its surface. The
spatially averaged pressure received from each
scattering element is proportional to F2(y,z) and
can be readily calculated by using the coordinate
system employed by O'Neil [9]. The total received
pressure is then obtained by integration over the
gated volume, v,^
p,^(co) = B(a,)yS(a3; x,y,z)F2(y,z)e-j2'^^/^dxdydz (3a)
^m
where B(u)) is
^-j2coR/c
B(aj) = (jwA/c)P(oj)^ (3b)
(2TrR)2
and the sub-index m indicates the m'-'' measurement.
As a first approximation for many tissues of
interest, S(u); x,y,z) characterizes weak scatter-
ing of identical, randomly distributed, isotropic
scattering elements. Under these conditions it
is convenient to factor the scattering function
as follows
The received signals are analyzed in the fre-
quency domain. The classic results obtained by
O'Neil [9] are applicable for the weakly focused
transducers used in this program. These results
show that, at a point Q in the focal zone, the in-
cident wave is essentially planar and its pressure
amplitude is given by
p-jkr
p' (oj) = jkA P{m)- F(r,e)
2TTr
(1)
where k = u/c is the wave number; u and c are
radian temporal frequency and propagation velocity,
respectively; P(a)) is the pressure amplitude at the
transducer surface with area A. The function
F(r,e) describes the beam pattern at a constant
range from the transducer and is equal to
2Ji(k a sin e)/(k a sin e) where Ji(6) represents
a Bessel function of the first kind and first
order with argument g.
It is useful to simplify this expression at the
outset. By virtue of the narrow angular beam-
S(w; x,y,z) = H(a))N(x,y,z)
(4)
Here H(aj) defines the reflectance characteristic
of a single scattering element while N(x,y,z)
describes their effective spatial concentration.
With eqs. (3) and (4), the total received pres-
sure in the m^'^ measurement can be written as a
Fourier transformation, with respect to axial range,
of the product of the gating function G(x) and a
function characterizing the weighted concentration
of tissue elements; i.e.
p^(w) = B(a))H(w) J G(x)|"
'N(x,y,z;
(5)
F2(y,z)dydz
e-j2(^x/c^^
The function G(x) describes the length of the range
gate and the range-weighting function (e.g. , Ham-
112
ming) applied in the measurement. Denoting the
Fourier Integral by Qfji(co), eq. (5) can be rewritten
as
p (w) = B(a))H(a))Q^((.
'^m m
(6)
It is important to notice the stochastic nature
of Qm(a)) resulting from the stochastic nature of
the tissue concentration function, N(x,y,z). Quali-
tatively, for equal, but non-overlapping tissue
volumes, Pm(u) can vary randomly from measurement
to measurement, as different tissue realizations
are illuminated with the ultrasonic beam. This
stochastic process can be meaningfully character-
ized in terms of the power spectrum, p, computed
by forming an ensemble average of the modulus-
square of Pni(a)) -, i ■ e. ,
<N(x,y,z)N(xi,yi,Zi)> = R.,(Ax,Ay,Az)
(9)
where Ax = x-xj. Ay = y-yi, az = z-zi. For this
case, substitution into eqs. (5) through (7) can
be carried out with considerable simplification.
The expected value of p, taken in the sense of an
average over the ensemble of tissue realizations
becomes
<P>
(10)
|B(aj)|:
|H(w)|2y J J Rg(Ax) Rp(Ay,Az)
P = 1 E |P (f^)
M ' ^m
m=l
(7a)
(7b)
m=l
In the present technique, the ensemble average
is computed from a series of observations of
|Pi,i(oj)[2 taken along adjacent, non-overlapping
beam headings within the examined tissue. The re-
sulting function p then constitutes an estimate of
the "true" power spectrum, <p>. (The symbol <•>
refers to ensemble averaging over all realizations
of the structure of interest.)
The statistical fluctuations associated with
estimates of <p> are of significant practical im-
portance. For tissues characterized by a
Gaussian-distributed concentration function and
a correlation volume much smaller than the sampling
volume, Vpi, Pni(a)) has a Gaussian distribution and
|Pni(u)p obeys Chi-square statistics. (Tissues
which possess many independent scattering centers
within Vm obey these statistics as well by virtue
of the law of large numbers.)
It follows from the properties of the Chi-square
distribution that independent measurements of
|P[ii(a))|2 exhibit significant variability; in fact,
at any value of u, the standard deviation, a, of
such measurements is equal to their mean value,
<p>. The relative fluctuations which persist after
averaging M independent realizations of |Pu,(a))|2
can be assessed from the ratio
<P> ± o
<P>
= 1 ± M"
18)
In spectral measurements, this degree of fluctua-
tion is evidenced at each value of u producing a
statistical "ripple" of the above magnitude about
the mean spectral shape. This result has been
employed to select a value of M that is suitable
for clinical measurements as discussed below.
B. Statistically Stationary Architectures
If the concentration function, N(x,y,z), is a
weakly stationary process, then its autocorrela-
tion function can be expressed as
R^(Ax,Ay,Az) x ^-^^^^^/^ d(Ax)d(Ay)d(Az)
where Rq and Rp are the autocorrelation functions
of the gate function, G, and directivity function,
F^, respectively.
For a limiting case that is of practical in-
terest here, the correlation volume defined by the
system-related functions Rq and Rp is much larger
than that of the tissue function Rn. Equation
(10) can then be approximated as
<P>
^ |B(a))|2 |H(w)|2 Rg(0) Rp(0,0) T(a3) (11)
where T(aj) depends only on the tissue structure
and is given by
T(w) ^ J f f Rf^(Ax,Ay,Az) d(Ay)d(Az
(12)
^-j2a.Ax/c
C. Spatial Distribution
of Rayleigh Scat<-erers
The above discussions are applicable directly
to tissues which can be modelled as a set of ran-
domly distributed, discrete Rayleigh scatterers:
i.e., scatterers whose sizes are much smaller than
the illumination wavelengths. We next treat the
case where the number of Rayleigh scatterers with-
in a unit volume is a Poisson process; i.e.,
K
N(x,y,z)
^ 6(x-x.) 6(y-y.) 6(z-z.:
i = l
(13)
where 6 is the Dirac delta function and (xi,yi,z-j)
are the random coordinates of the ith scatterer
which are uniformly distributed over the inspected
tissue volume. K is the number of scatterers per
unit volume and obeys Poisson statistics with para-
meter <K>. It follows [10] that R^, the autocor-
relation function of N, is
= <K> 6(Ax) 6(Ay) 6(Az) + <K>2
(14)
113
Using eq. (14) for in eq. (10), one obtains
<P> = |B(a))|2 |H(to)|2 <K>Rg(0)Rp(0,0) + (15)
<K>2 |G(a)) |2 / / Rj-(Ay,Az) d(Ay)d(Az)
where G(u)) is the Fourier transform of Rg(ax).
The first term in eq. (15), proportional to
the expected number of scatterers, is identi-
fied with non-coherent scattering. The second
term, proportional to the square of the ex-
pected number of scatterers, is identified
with coherent scattering. The relative con-
tributions of both components to the total mea-
surement varies with frequency. The non-
coherent spectral component is distributed over
the frequency domain, whereas the coherent com-
ponent primarily contributes at zero frequency
providing that the power spectrum |G(u))|2 can
be adequately approximated as a Dirac delta
function at the frequency origin.
In practice, non-coherent spectra predominate
in clinical observations. This occurs because
spectral observations are carried out at high
frequencies and a time-weighting function is used
to suppress the sidelobes of |G(a))|2. These side-
lobes fall-off at a rate which is more rapid than
the a)"2 characteristic associated with a rectangu-
lar gating function.
Equation (15) can be used to address the topic
of spectral normalization. The non-coherent term
is proportional to |H(a3)|2 which for Rayleigh scat-
terers has an aj'+ dependence. However, other
system-related terms are also frequency-dependent.
To account for these factors, spectral data are
normalized using calibration spectra obtained from
a flat, water-glass interface viewed at normal in-
cidence and located at R + L/2. It can be shown
that this normalization removes the system-related
frequency dependence in the non-coherent component
of eq. (15) and that the normalized spectrum is
indeed proportional to u'*. Experimental results
verifying this conclusion are presented in a sub-
sequent section.
These results and those of preceding sections
are applicable to a variety of tissue structures.
However, all tissues of interest do not fall with-
in simple categories and further analysis is war-
ranted to treat more complex tissue structures
and to investigate the most appropriate approaches
to data normalization. Such analyses are being
conducted as the clinical data base is expanded
and histologic preparations become available.
3. Spectrum Analysis System
The spectrum analysis techniques described above
have been implemented with an on-line clinical sys-
tem. The system is integrated with a high-
resolution A- and B-scan instrument which is em-
ployed routinely in ophthalmic examinations (fig.
2). The operation of the system is described
elsewhere [6] and is only briefly summarized here.
The tissue segment to be analyzed is selected by
observing A- and B-scan displays in which the posi-
tion of the moveable, system range gate is super-
imposed. These displays are also monitored to in-
sure that the beam does not traverse the absorp-
tive ocular lens and that the tissue segment to be
analyzed lies within the transducer's focal zone.
After the range gate has been placed over the
desired tissue segment, processing is initiated.
First, the gated rf echo complex is multiplied by
a time-weighting function to suppress spectral side
lobes which would prevent accurate measurements of
steeply rising or falling spectra. Then, the
gated signals are applied to a scanning electronic
spectrum analyzer which computes the desired spec-
trum comprised of 50 spectral elements occupying
adjacent 300-KHz frequency bands.
An ensemble of these spectra are computed along
13 adjacent, non-overlapping beam headings and
entered into an on-line averager. The spectral
ensemble is obtained by executing a slow sector
scan during which an optical encoder initiates a
spectrum analysis computation each time the trans-
ELECTRONIC
PULSER
T/R
TIME-WEIGHTED
SCANNING
SPECTRUM
RANGE GATE
— a»
SPECTRUM
SWITCH
UNIT
ANALYZER
AVERAGER
SECTOR SCAN
ANALYSIS REGION
(1.5 X 4 0 mm )
A/B -SCAN
MONI TORS
STRI P
CHART
RECORDER
Fig. 2. Configuration of spectrum analysis system.
114
ducer heading is incremented by one angular beam-
width (0.3 degrees at 10 MHz). (In practice, 100
partly redundant spectra are averaged to improve
electronic signal-to-noise ratios.) Averaged
spectra are presented on a strip chart recorder
which also displays frequency and amplitude cali-
bration signals for use in subsequent processing.
After each examination session, calibration
spectra are recorded from an optically flat glass
plate situated at each of the ranges used in tis-
sue measurements. The calibration spectra and
tissue spectra are subsequently digitized and
entered into a computer for spectral normalization,
smoothing, regression analysis, and data catalogu-
ing.
Table 1 lists the salient system parameters
which influence the detail with which tissue char-
acteristics such as |H([jj)|2 and T(aj) can be examin-
ed. As discussed below, the frequency coverage,
spectral resolution, and number of non-overlapping
beam positions are of central importance in tissue
studies.
Table 1. Nominal system parameters.
Frequency coverage 5 to 13 MHz
Spectral resolution
Gating function
Spectrum analyzer
Gated range segment
Lateral scan dimension
Beam width (10 MHz)
0.6 MHz
0.3 MHz
1. 5 mm
4 mm
0. 3 mm
Number of distinct beam positions 13
Dynamic range 30 to 35
The frequency coverage achievable in spectral
measurements has been found to approximate the
20-dB system bandwidth as defined by glass plate
spectra. This coverage is nominally 5 to 13 MHz
for a broadband transducer with a 10 MHz resonant
frequency. Beyond this range, limited signal-to-
noise ratios are encountered and spectral normali-
zation can become inaccurate. The available fre-
quency coverage determines the spectral charac-
teristics of tissue elements with specific scale
sizes. Within the above frequency range, Rayleigh
scattering is encountered for spherical elements
whose diameters are less than 35 ym. As discussed
below, membranes, or periodic features, with axial
dimensions larger than 100 pm will produce period-
ic spectra with at least one spectral cycle dis-
played over the achievable 8 MHz coverage. In ad-
dition, tissue septa whose thicknesses are less
than 30 ym will exhibit monotonical ly rising
spectra.
Spectral resolution is determined by the 2-us
gate duration and the weighting function. These
variables have been chosen to achieve a 0.6 MHz
resolution while the spectrum analyzer employs a
0.3 MHz filter bandwidth. The short gate duration
corresponds to a 1.5 mm tissue depth permitting
analysis within small ocular tumors and avoiding
significant signal weighting due to attenuation
within the gated echo complex.
The number of independent spectra which are
ensemble averaged is an important factor in assess-
ing the residual statistical fluctuation with
estimations of mean spectral shape and amplitude.
Usually, 13 spectra from non-overlapping beam
positions are used to form the ensemble average.
For the conditions described with reference to eq.
(8), a statistical spectral "ripple" of approxi-
mately ± 1 dB is expected. While spectral fluc-
tuations are often confined within this range,
there are situations where significantly larger ex-
cursions occur suggesting that the analyzed tis-
sues exhibit relatively large correlation volumes.
(Further analyses of these parameters in terms
of the accuracies of spectral slope and attenuation
estimations have been presented in a previous
publ ication [7] . )
4. Representative Results
Before clinical deployment, the performance of
the spectrum analysis system and the applicability
of the theoretical approach was verified on a num-
ber of test targets. One type of test target was
a thin plastic membrane immersed in distilled water
and viewed at normal incidence. For this target,
N(x,y,z) is proportional to 6(x-Xq) - 6(x-Xq-D)
where D represents thickness and x^ locates the
proximal membrane surface. From eq. (5) and (7),
the calculated power spectrum, is proportional to
sin (2TTfD/c) so that spectral minima will occur
at frequencies separated by an interval equal to
c/2D. The observed spectrum, shown in figure 3,
exhibits a scalloped shape in agreement with this
result. The measured frequency interval between
successive spectral minima is 3.6 MHz which cor-
responds to a calculated thickness of 360 pm.
This value is within 5 percent of the measured
membrane thickness.
Spectral data were also obtained for dilute
solutions of plastic microspheres with 25 ym
diameters. These dilute solutions insure that
single scattering from points is approximated over
the spectral measurement band. Moreover, it is
reasonable to assume that their spatial distribu-
tion forms a Poisson process since the random posi-
tion of each scatterer should be independent and
uniformly distributed within the analyzed volume.
Accordingly, measured spectra should be described
by eq. (15).
Figure 4 shows normalized spectra measured for
two microsphere concentrations. The spectral
curves agree very closely with the u"* dependence
expected from the normalized, non-coherent spectral
component in eq. (15). In addition, a fourfold
change in microsphere concentration was found to
produce a proportional 6-dB change in spectral
amplitude. This amplitude change is consistent
with non-coherent scattering, in which received
spectral power is proportional to the mean number
of scattering particles.
Clinical measurements have been carried out on
a wide variety of normal and diseased structures
in the eye and orbit. In vivo data on more than
200 cases have been acquired and digitally cata-
logued as part of an on-going research effort. The
following paragraphs review some of the observed
spectral features which relate directly to cases
treated in the preceding discussions of analytic
and experimental results.
115
-20
-24
^ -28
u -32
Q
3
-36
-40
-44
-48
-5 2 -
Af = 3.6 MH z
D = 360
J L
J I I L
J I L
5 10
FREQUENCY (MHz)
Fig. 3. Spectrum of echoes from plastic membrane.
Oi-
- - 5
-10
-20
MICROSPHERE SPECTRA
25 ^im DIAMETER
FULL CONCENTRATION
1.7 X 10^/cm
RAYLEIGH
RESPONSE (f^*)
ONE-QUARTER CONCENTRATION
0 43 X 10^ /cm^
0.42
l_
J L
Fig. 4.
7 8 9 10 1 1 12 13 14 15 16 17 18
FREQUENCY (MHz)
Spectra obtained for two concentrations
of dilute microsphere suspensions.
Detached retinas, observed prior to the develop-
ment of gross degenerative processes, constitute
thin membranes bounded, anteriorly, by vitreous
humor and, posteriorly, by fluid exudate. Spectra
obtained from these structures display a marked
scalloped appearance similar to that seen for the
plastic membrane (fig. 3). This fact is demon-
strated in figure 5 which shows clinical spectra
in decibels relative to a glass ^'late (dBr) ob-
tained from three different cases of retinal de-
tachments. Retinal thickness can be computed from
the frequency repetition interval of these spectra.
Using a nominal propagation velocity of 1.5 mm/ps
for the retina, the computed values in the illus-
116
Fig. 6. Clinical spectra obtained for vitreous hemorrahages (9 cases). Spectral amplitudes
have been off-set by indicated values.
117
trated cases are on the order of 190 ym and are
consistent with expected retinal dimensions.
Vitreous hemorrhages have been studied exten-
sively with the spectrum analysis system. In their
unorganized state, these hemorrhages contain ran-
domly placed aggregations of hemorrhagic debris
dispersed in sections of the vitreous humor. Un-
organized hemorrhages consistently produce spectra
whose amplitudes increase markedly with increasing
frequency as shown in figure 6. The rate of in-
crease can sometimes be as rapid as the f"* charac-
teristic associated with Rayleigh scattering:
this fact indicates scattering from a uniform
distribution of elements that are smaller than
35 ym. Spectral amplitudes are typically -70 dBr
near 5 MHz: these very low levels can often be
exceeded by returns from isolated specular reflec-
tors or, in the extreme, by the system noise level.
At higher frequencies, the spectral amplitudes
produced by hemorrhagic debris are significantly
larger and exceed these background levels.
Occasionally, other spectral shapes have been
observed within an examined vitreous hemorrhage.
They indicate a possible high-density packing or
organization of the hemorrhagic debris. The
presence of membranes has also been found to af-
fect the observed spectral shape. For example,
figure 7 shows the scalloped spectra received from
a range-gated volume which contained both hemor-
rhagic debris and a thickened intravitreal
membrane.
5. Conclusion
The analysis outlined in the preceding sections
has proven extremely useful in designing the spec-
trum analysis system and in selecting system para-
meters, such as gate dimensions, for clinical ap-
plication. It has also formed a constructive
framework for interpreting in vivo data, especial-
ly for detached retinas, vitreous hemorrhages, and
intra-vitreal dispersions of cholesterol aggrega-
tions.
The analysis is being extended to accommodate
more complex tissue architectures. This effort
is proceeding using clinically measured spectra
and subsequent histologic preparations when avail-
able. Of particular interest are intra-ocular
tumors such as malignant melanomas and metastatic
carcinomas.
Other topics under investigation include the
use of spectral data to estimate tissue attenua-
tion characteristics. Presently, attenuation
estimates are formed from ratios of power spectra
measured at sequential range sites separated by
aR. If the examined tissue architecture satisfies
certain conditions (e.g. , statistical homogeneity),
then such ratios describe the attenuation ex-
perienced within AR. Spectral data indicate that
this approach is appropriate for some tissues
(e.g. , orbital fat) but is not applicable to other
tissues (e.g. , heterogeneous tumors). In these
latter cases, spectral ratios might serve as a
useful index of statistical homogeneity.
As the clinical data base is expanded, these
investigations can proceed toward the ultimate
goal of applying spectrum analysis techniques to
supplement conventional A- and B-scan ultrasono-
graphy.
Acknowledgments
The authors wish to acknowledge the collabora-
tion of D. Jackson Coleman and Louise Franzen in
the clinical aspects of this work. They also wish
to thank Angel Rosado for his assistance in data
processi ng.
Portions of this work were supported by Public
Health Service Grants EY-0I212-04 and EY-01218-04
from the National Eye Institute.
-44 -
m
u -52
Q
K -56
^ -60
<
-64
-68
^-72
CO
-76
PATIENT 3943
10
FREQUENCY (MHz)
15
Fig. 7. Clinical spectrum obtained for a vitreous hemorrhage and
associated intravitreal membrane.
118
References
[1] Waag, R. , Lerner, R. , and Gramiak, R. ,
Swept-Frequency Ultrasonic Determination of
Tissue Macrostructure , in Ultrasonic Tissue
Characterization, M. Linzer, ed. National
Bureau of Standards Spec. Publ . 453, pp.
213-230 (U.S. Government Printing Office,
Washington, D.C. , 1976). ■ • ■
[2] Lele, P. and Namery, J., A Computer-Based
Ultrasonic System for the Detection and
Mapping of Myocardial Infarcts, in Proceed-
ings of the San Diego Biomedical Symp. 13,
121-132 (1974).
[3] Sigelmann, R. and Reid, J., Analysis and
measurement of ultrasound backscattering
from an ensemble of scatterers excited by
sine-wave bursts, J. Acoust. Soc. Am. 5^,
1351-1355 (1973).
[4] Lizzi, F. L., St. Louis, L. , and Coleman,
D. J., Applications of spectral analysis in
medical ultrasonography. Ultrasonics 14 (2),
77-80 (1976).
[5] Lizzi, F. , Burt, W. , and Coleman, D. J.,
Effects of ocular structures on the propaga-
tion of ultrasound in the eye. Arch. Ophthal-
mol . 84, 635-640 (1970).
[6] Lizzi, F., Laviola, M. , and Coleman, D. J.,
Ultrasonic Tissue Characterization Using
Spectrum Analysis, in Proceedings of SPIE/SPSE
Conference on Application of Optical In-
strumentation in Medicine V, Society of
Photo-Optical Instrumentation Engineers,
Palos Verdes Estates, California, pp. 322-
328 (1976).
[7] Lizzi, F. , Laviola, M., and Coleman, D. J.,
Tissue Signature Characterization Utilizing
Frequency Domain Analysis, in Proceedings of
IEEE Ultrasonics Symposium 1976, Institute
of Electrical and Electronics Engineers, Inc.,
New York, pp. 714-719 (1976).
[8] Lizzi, F. and Elbaum, M. , Ultrasonic Spectrum
Analysis Techniques in Medical Ultrasound,
Riverside Research Institute Technical Re-
port, Riverside Research Institute, New York,
N.Y. (in preparation).
[9] O'Neil, H., The theory of focussing radiators,
J. Acoust. Soc. Am. 21_, 516-526 (1949).
[10] Pratt, W. Laser Communications Systems,
pp. 255-262 (John Wiley and Sons, New York,
1969).
119
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed. , National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
TISSUE CHARACTERIZATION IN VIVO BY DIFFERENTIAL ATTENUATION MEASUREMENTS
Salavatore Levi and Jose Keuwez
Ultrasound Lab and F.R.E.S.E.R.H. , Hopital Brugmann
Universite Libre de Bruxelles, Brussels, Belgium
A practical ultrasound method has been applied for the differentiation in vivo
of pelvic tumors. This method has the advantage of being applicable during usual
ultrasonic examination and does not require information about the tissues involved,
nor the surrounding tissues.
The method is based upon the fact that the ultrasonic attenuation increases with
frequency; the comparison of the echoes from the posterior and anterior boundaries
of the tumor leads to a coefficient of differential attenuation. This coefficient
has been determined in examinations of 95 gynecological patients; the organs and
tumors explored were normal uteri, leiomyomas and ovarian cysts. The values of the
coefficient of differential attenuation obtained for these three categories of
tissue are sufficiently separated, which enables us to make a differentiation.
Key words: Characterization; differential attenuation; tumor; ultrasound.
1. Introduction
Ultrasound has been shown to be effective in
the diagnosis of tumors and is generally capable
of differentiating between solid and liquid com-
ponents. The current acoustic methods, which
have been applied for a number of years, give de-
pendable results in 93 percent of the leiomyoma
cases and in 84 percent of the cases involving
ovarian cysts [1 ]^ .
Ultrasonic imaging has reached a high level
of accuracy, especially with the application of
"grey-scale" techniques. However, other aspects
of diagnostic ultrasound have not yet reached
their full potential. More precise information
on the viscoelastic properties of tissues, espe-
cially those of tumors, should be the next step
in research on diagnostic methods.
Quantitative studies should produce more mean-
ingful numerical data regarding the various types
of tumors investigated. Such data will be rep-
resentative of the acoustical properties of the
tissues and complete the subjective data produced
by imaging.
One of the most important of these properties
is the attenuation of sound. The loss in acoustic
intensity in travelling through the biological
tissues is caused mainly by absorption and scat-
tering. These causes may be grouped under the
term "attenuation". It is difficult to assess the
individual contribution of these mechanisms (ab-
sorption, scattering, ...) and it is therefore
quite preferable to confine oneself to examining
attenuation as a whole.
The importance of the various mechanisms is de-
pendent on the wave frequency; therefore the at-
^Figures in brackets indicate literature
references at the end of this paper.
tenuation is a function of frequency. The attenua-
tion is a characteristic of a tissue, but general-
ly it is quite impossible to measure it in vivo
without invasive techniques. We chose the deter-
mination of the variation of the attenuation with
frequency as a basis for the differentiation of
tissues [2-4]. A similar concept has been pro-
posed by Kossoff [5,6], but no results have been
published. For a good understanding, we envision-
ed a model similar to that published by Hill [7].
2. Model and Analysis
The following analysis is approximate and
does not take into account various mechanisms
which could be important. The objective of this
analysis is to give an idea of the relation be-
tween the ultrasonic properties of tissues under
investigation and the coefficient of differential
attenuation which will be defined below.
Generally any mass^ consists of several layers
of tissue which have their own properties; for
example, the uterine fibroma in which a layer of
normal muscle surrounds pathological tissues.
The model (fig. 1) has been constructed in order
to represent the different layers of the region
under investigation and of the mass. Each layer
is characterized by an acoustic absorption coeffi-
cient, and each interface by a back-scattering
cross-section (a), (o is used here to take account
of all the mechanisms which contribute to the loss
of energy at this interface.)
These layers are separated into two sets; p
layers between the transducer and the mass, the
mass being constituted of r - p = n layers. In
each layer, one of the important mechanisms in
2ln this paper, the term "mass" will include
the normal uterus, leiomyoma, and cyst.
121
MASS
ai
V 1
p layers
n layers
Fig. 1. Model of the different layers of the
region under investigation.
which acoustic energy is degraded as heat by in-
ternal friction of viscosity is the absorption.
This loss of acoustic energy follows the relation
given by the equation:
A = Aq exp(-ax)
(1)
where a is the amplitude absorption coefficient,
Aq and A are, respectively, the wave amplitude at
some reference point and the amplitude at a fur-
ther distance, x, from that reference point.
Let us consider a sound wave propagating from
the source S to the point R (fig. 1). The ampli-
tude of that wave at R is given by
A(R,v) = Ao(v) o(R,v) exp(-ai(v)<Si) (2)
. . . exp(-ap(v)6p)
where v is the wave frequency, a-j and s-j are, re-
spectively, the amplitude absorption coefficient
and the thickness of the layer i; a{\>) is a factor
which describes the decrease of amplitude at the
several interfaces. The amplitude, SA, of the
reflected wave coming from R on to the transducer
is related as follows:
SA(R,v) = AqIv) a(R,v) exp(-2 ^ cx^^i) (3)
i = l
If (vi,v2) are two different frequencies, let us
consider the ratio of the amplitudes SA(R,vi) and
SA(R,V2); according to eq. (3), this ratio is given
by the following equation:
SA(P,Vi) Ao(vi) o(R,vi)
In = In + In
SA(R,V2) Ao{v2) a(R,V2)
r
- 2 ^ (ai(vi)-ai(v2))
i = l
Similarly, (see fig. 1)
SA(P,Vi) Ao(vi) o(P,Vi)
In = In + In
SA(P,V2) Ao(V2) ct(P,v2)
2 (ai (vi )-aT (v2) )
i=l
(4)
(5)
Now we introduce J as the difference of eqs. (4)
and (5).
SA(R,vi) SA(P,Vi)
J = In In
SA(P,V2)
a(R,V2)
J = In In ■
a(P,Vi) o(P,V2)
2 J2 (ai (vi (V2) )6i
i=p+l
(6)
(7)
Equation (6) can be written in the following form:
(n(vi) - n(v2)) - (a(vi) - a(v2))
2d
= Af) - Aa
where
a^6] = ad (d = thickness of the mass) (10)
i=p+l
In
t(R,n
a(P,v]
2dn(v)
(IT)
Examining eq. (8), we see that J/2d depends only
on the ultrasonic properties of the layers com-
prised between P and R; J/2d is a characteristic
of the mass and tells us about the variation of
attenuation with frequency.
Let us remark that, for the calculation of
the value of J, we only need the knowledge of
four amplitudes, and no information is required
about the tissues involved nor the surrounding
tissues.
3. Materials and Methods
We have defined eq. (6)
J = In
SA(R,Vi)
SA(R,V2;
In
SA(P,Vi;
SA(P,V2;
■ (12)
which is the basic equation for characterization
based upon differential attenuation. For the cal-
culation of J it is necessary to measure the
amplitudes of the waves reflected by the anterior
and posterior boundaries of the mass. This mea-
surement must be performed at two frequencies
(vi and V2). Theoretically, it is possible to
undertake these measurements with a single pulse
[6], however, we have chosen to adapt eq. (10) by
considering that these amplitudes are for waves
generated by two relatively narrow band (9.5 MHz)
transducers .
These amplitudes have been measured with the
ultrasonic equipment "Kretz Combison I" [2-4].
For each amplitude measurement, all the settings
were identical except that of the master gain.
This last one was adequately adjusted to obtain an
amplitude of given level on the A-Scope for the
echo of interest (I). The corresponding reading
122
L(I,v) (in dB) was used in the following equation
based on eq. (12)
Table 1. Statistical analysis.
2d
L(R,Vi) - L(R,V2) - (L(P,Vi) - L(P,V2)) (13)
where d = dimension of the mass. We call y the
coefficient of differential attenuation. This at-
tenuation coefficient has been determined in exami-
ing 95 gynecological patients. The masses under in-
vestigation were the normal uterus, leiomyoma and
cyst. The nominal frequencies were 2 MHz and 4 MHz.
4. Results -
At present, twelve cases have been labeled as
normal uterus, and 34 other cases have been clas-
sified as leiomyoma or cyst after surgery. We
plotted the values of the differential attenuation
coefficient, y, versus the dimension, d, of the
mass (fig. 2). The plot shows that y is not in-
fluenced by d and that the y belonging to the
three categories of tissues (normal uterus, leiomy-
oma and cyst) are quite characteristic. Some
cysts have shown an abnormally high differential
attenuation for masses with liquid content; we also
obtained negative values for some cysts, however,
this peculiarity has been reported and explained
[8]. The values of y for the normal uterus are
higher than those obtained for the leiomyoma, which
is in agreement with the ultrasonic properties of
these tissues and with previous results [4].
The results of the statistical analysis are shown
in table 1.
y (dB/cm)
1-8.
1.6.
1.4.
1.2.
1.0.
0.8.
0.6.
0.4.
0.2.
0.0
-0.2
+ normal uterus
0 leiomyoma
X cyst
O OOCP
d (cm)
12
15
Fig.
2. Plot of the values of the differential
attenuation coefficient, y, versus the
dimension, d, of the mass.
Organ
No.
Mean
y(db/cm)
S.D.
Normal uterus
Leiomyoma
Cyst
12
22
10
1.39
0.57
0.13
0.27
0.22
0.20
The absolute values are not interesting by them-
selves because y is also depending on the instru-
mentation.
5. Comments and Discussion
If in vitro it is possible to imagine trans-
mission methods, it is more difficult in vivo
where generally only reflections methods are ap-
plicable. Unfortunately, it is not enough to know
the amplitude of the reflected signals to tell
something about the tissues involved. It is also
necessary to know the properties and the orienta-
tion of the interfaces which have partially re-
flected the sound waves.
When working in vitro, the samples can general-
ly be chosen, cut and positioned as desired. In
vi vo , many constraints have to be taken into ac-
count. Theoretically our method avoids some of
them; for instance no information concerning the
surrounding tissues is required. However, diffi-
culties due to the orientation of the interfaces
remain. Sometimes it was difficult to obtain a
posterior echo because of the orientation of the
tumor with respect to the abdominal wall. The
orientation is also affected by the cardio-
respiratory movements. Besides the difficulties
of measurement, one should be aware that the re-
fraction properties are frequency dependent. This
could cause important differences in amplitude
when using two frequencies vj and V2 at oblique
incidence. The dispersion of our results can in
part be explained by this last discussion as well
as by considering effects due to the beam geometry.
We characterized our transducers [4]; however
this characterization was made in a nonattenuating
medium (water). In our experiments the sound was
propagating into an inhomogeneous and attenuating
medium (for instance the abdominal wall) which
could strongly influence the beam geometry.
Therefore it is illusory to pursue this discussion
based on our transducer characterization. How-
ever one must be aware that the beam geometry can
dominate the results. That is probably the reason
we did not obtain coherent results when working
with other transducers. This method, as it is ap-
plied, considerably extends the duration of the
examination, which is generally performed with a
full bladder. Under these circumstances it is
difficult for the patient to lie still. Techni-
cal improvements are necessary.
6. Conclusion
The characterization of tissues based upon the
variation of attenuation with frequency seems to
be suitable and helpful in cases where scans alone
do not permit an accurate diagnosis. However, we
would like to emphasize that it was not always
possible to determine the coefficient of differen-
tial attenuation, for several reasons developed
123
in the previous paragraph. The results reported
in this paper are those obtained for frequencies
of 4 and 2 MHz; however, we also determined this
coefficient for frequencies of 2 and 1 MHz. The
results obtained in this last case were incoherent
and the overlap between the three categories of
tumors investigated was important.
The various inconveniences described above
should be resolved by a better choice of trans-
ducers and the way of determining the various
amplitudes required for the computation of the
coefficient of differential attenuation.
Acknowledgments
We wish to thank Professor R. C. Eggleton
(Indianapolis Center for Advanced Research,
Indianapolis, Indiana) for helpful suggestions
in preparing this paper, and Mrs. J. Drake
for her secretarial assistance.
References
[1] Levi, S. and Delval, R., Value of ultrasonic
diagnosis of gynecological tumors in 370
surgical cases. Acta Qbstet Gynecol Scand 55
261-266 (1976).
[2]
[3]
in Abstracts of the 2nd World Congress on
Ul trasonics in Medicine, p. 81 (Junge und
Sohn, Muchen, 1975).
[4] Levi, S. and Keuwez, J., An Attempt to Find
a Differential Attenuation Coefficient for
Ultrasonic Diagnosis of Pelvic Tumors in
Vivo, in Ultrasound in Medicine, Vol. 3B,
p. 1989 (Plenum Press, New York, 1977).
[5] Kossoff, G., Display techniques in ultrasound
pulse echo investigations; a review, J. Clin.
Ultrasound 2, 61 (1974).
[6] Kossoff, G., Reflection Techniques for Meas-
urements of Attenuation and Velocity, in
Ultrasonic Tissue Characterization, M. Linzer,
ed., National Bureau of Standards Spec. Publ .
453, pp. 135-139 (U.S. Government Printing
Office, Washington, D.C., 1976).
[7] Hill, C. R., Echoes from Human Tissues, in
Ultrasonics International '75 Conference
Proceedings, 24-26 March 1975, Imperial
College, London, pp. 20-22 (Guildford IPC
Science and Technology Press Ltd., London,
1975).
Levi, S., Essai d'analyse quantitative des [8]
echogrammes de tumeurs pelviennes; Journees
d^Etudes sur les Ultrasons Appliques a la
Medicine, Nancy (1974), Resume des Communi-
cations, p. 40.
Levi, S., An Attempt to Differentiate Pelvic
Tumors in Vivo by Attenuation Measurements,
LeCroissette, D. H. and Heyser, R. C,
Attenuation and Velocity Measurements in
Tissue Using Time Delay Spectrometry, in
Ultrasonic Tissue Characterization, M. Linzer,
ed., National Bureau of Standards Spec. Publ.
453, pp. 81-95 (U.S. Government Printing
Office, Washington, D.C., 1976).
124
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed.. National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C. , 1979).
STATISTICAL ESTIMATION OF THE ACOUSTIC ATTENUATION COEFFICIENT SLOPE
FOR LIVER TISSUE FROM REFLECTED ULTRASONIC SIGNALS
Roman Kuc,^ Mischa Schwartz, ^ Nathaniel Finby,^ and Frank Dain^
^Department of Electrical Engineering and Computer Science
Columbia University
New York, New York 10027, U.S.A.
^Department of Radiology
St. Luke's Hospital Center
New York, New York 10027, U.S.A.
The acoustic attenuation coefficient measured in dB/cm is known to increase linear-
ly with frequency for liver tissue. The slope of this linear function, denoted by b,
has been shown by other investigators to be an indicator of tissue state. 6 is usual-
ly measured from the change in the log spectrum experienced by an acoustic pulse when
it is transmitted through the tissue. This paper presents research in estimating 6
from the reflected signals which are currently used in the clinical environment to
generate diagnostic images. The reflected signals from internal tissue structures
(in liver - the vascular and biliary systems) are distorted by the irregular reflec-
tor shapes. By modeling the roughness of the typical acoustic reflector, the dis-
tribution of the received spectra can be derived and the 95 percent confidence limits
for the measured B can be calculated. The confidence limits are presented in terms
of the tissue size and average reflector density. Experimental results using the
reflected signals from in vitro refrigerated and formalin-fixed liver sections are
presented.
Key words: Computer processing; estimation theory; liver attenuation; spectral
analysis; statistical modeling; ultrasonic tissue characterization.
1. Introduction
As an acoustic pulse propagates through liver
tissue it experiences a frequency dependent attenu-
ation. Pauly and Schwan [l]i observed that the
acoustic attenuation was a linear function of fre-
quency and postulated that this behavior was primar-
ily the result of macromolecular relaxation pro-
cesses. Lele Gt_ aj_. [2] confirmed the linear fre-
quency dependence of the acoustic attenuation and
in addition experimentally demonstrated that the
slope of the linear dependence increased for necro-
tized tissue. Lele's experiments were performed by
transmitting an acoustic pulse of known shape
through a thin liver section and observing the
change in the log power spectrum as the pulse propa-
gated through the tissue. In this paper we discuss
our work in estimating the slope of the linear fre-
quency dependent attenuation, which we denote by
the coefficient B, from the reflected acoustic
signals, typical of those currently used in the
clinical environment to generate the acoustic
images.
The reflections observed from liver tissue are
caused primarily by the vascular and biliary sys-
^Figures in brackets indicate literature
references at the end of this paper.
tems that permeate the liver volume. Because the
typical reflector encountered within the liver is
neither plane nor normally incident to the inci-
dent pulse, the log spectra of the reflected sig-
nals are distorted. In this paper we account for
the distortion caused by the irregularly shaped
reflector by modeling the reflector as a random
linear filter having an impulse response which is
a sample function from a zero mean, white Gaussian
process. From this assumed model we can derive
the distribution of the observed sample spectrum.
Using this distribution we can calculate confidence
intervals for the resulting estimates of e. Final-
ly, we predict the sensitivity of the estimator on
the tissue thickness, L, by assuming that the liver
tissue has a reflector density y reflectors per cm.
Experimental results using in vitro refrigerated
and fixed liver tissue specimens will be compared
to the analytically predicted performance.
2. Acoustic Attenuation of Liver Tissue
We first describe the experimental procedure
that we use to measure g from transmitted pulses.
This is used to verify the reflection measurements.
The power spectrum of the incident pulse, denoted
by Sx(f) in figure 1, is found by reflecting the
pulse from a plane, normally incident reflector
125
LIVER
d
Power Spectrum
dB
The average values of 6 that we obtained for in
vitro refrigerated and fixed liver specimens are
shown in figure 1. These values agree with those
found by other researchers [1,2]. The transmis-
sion technique works well when thin, parallel cut
sections of tissue are used. When large hetero-
geneous tissue specimens, or tissue specimens
which have irregular shapes are used, the propa-
gating pulse becomes distorted producing artifacts
in the measurements [3,4].
We will extend this technique of measuring 6 to
signals reflected from tissue structures by sub-
tracting the log spectra of signals from two re-
flectors within the tissue. But because of the
distortion to the spectra caused by the irregular
reflectors the observed spectral difference will
be a distorted linear function. The coefficient 6
will be estimated by fitting a straight line to the
spectral difference and dividing by the round-trip
distance between the reflector locations. In order
to find a suitable estimation technique and the
confidence intervals for the resulting estimate we
will derive the distribution of the distortion to
the log spectrum in the next section.
3. Reflector Model
FIXED LIVER
(e = 1.20 dB/cm-MHz)
REFRIGERATED LIVER
(e = 0.63 dB/cm-MHz)
f (MHz)
Fig. 1. Transmission experimental configuration
results. An acoustic pulse propagates
through liver sample having thickness d,
is reflected by plane, normally incident
reflector and propagates through sample
a second time. The liver acts like a low
pass filter, attenuating the high fre-
quencies in the power spectrum. The dif-
ference between the power spectra is found
to be linear with a slope which is an in-
dicator of tissue state. The parameter 6
is the observed slope divided of twice the
tissue thickness.
♦
with no tissue in the acoustic path. The power
spectrum of the pulse transmitted through the tis-
sue of thickness d back to the transducer is de-
noted by SY(f). The effect of the tissue acoustic
attenuation can be found by taking the log spectral
difference (in dB) of the two spectra. For a line-
arly frequency dependent acoustic attenuation ibhis
will be equal to
10 logS„(f) - logs (f) = 2 3df
(1)
The value of B, in units of dB/cm-MHz, for a given
tissue can be measured by dividing the observed
slope of the log spectral difference by the acous-
tic path length through the tissue. For reflected
signals the acoustic path length is equal to twice
the tissue thickness.
We propose to model the irregular reflector as
a random linear filter with an impulse response,
denoted by r(n) for n = 1, 2, KTj, which is
sample function from a zero mean, white Gaussian
process. The probability density function (p) of
r(n) is then given by
p(r(n)) =
1
exp
r2(n]
(2)
For a white Gaussian process, the values of r(n)
and r(m) for sampling times n m are independent.
This appears to be a reasonable assumption in this
work, considering the irregularity of the reflect-
ing vascular structures [5]. This is verified by
the experimental results to be described later.
If Sx(f) is the power spectrum of the acoustic
pulse incident to the reflector having the impulse
response r(n), then the power spectrum of the re-
flected signal, SR(f), can be written as
SR(f) = |R(f)|2Sx(f)
(3)
where |R(f)|2 is the power transfer function of the
reflector [6].
We will estimate 6 from the log spectral dif-
ference. The logarithm (in units of dB) of eq. (3)
is equal to
10 logSj^(f) = 10 logS^(f) + 10 log |R(f)|2
= 10 logS^(f) +
n(f)
(4)
(5)
The last equation indicates that the effect of
an irregular reflector is an additive distortion to
the log spectrum of the incident pulse at the re-
flector location. Furthermore, since the reflector
impulse response is assumed to be zero mean, white
Gaussian, it can be shown [6] that the random
variables
iR{fi)r
(6)
126
are mutually independent and have an exponential
distribution (the chi-square distribution with two
degrees of freedom), at frequencies f^ which are
separated by the fundamental frequency, F = l/KIg.
Since we are seeking the distribution of the
distortion in the log spectral domain (in dB) we
set
n. = 10 logC. = 10 log |R(f.:
(7)
Using the relation for the pdf of a transformed
random variable [7,8] the pdf of the distortion to
the log spectrum n-,- is found to be equal to
p(n ) = _
^ 4.34 a^T
exp
r s
"i
4.34
n . "
1
.4.34
a2T
r s
(8)
for
'1
This distribution is shown in
figure 2a.
If we let h = In a^Ts, then a^Tj = e^
can then be written
P(n^.)
4.34
exp
n .
1
4.34
- h
4.34
This last result shows that the effect
Eq. (8)
(9)
the vari-
ance of the reflector impulse response Oy on the
distribution of the log spectral distortion is only
a shift by an amount h.
We shall now describe the technique we have de-
veloped to partition the total reflected signal
from a liver tissue into segments which are used
for calculating the sample spectra of the returns
from the irregular reflectors.
4. Processing of Reflected Data
An acoustic pulse having an arbitrary power
spectrum Sx(f) propagates into a tissue medium
and reflections due to changes in tissue acoustic
impedance are detected by the transducer. A typi-
cal reflected signal denoted by y(n), is shown in
figure 3. The envelope of y(n) is calculated by
rectifying the sample values and low pass filter-
ing with a digital nonrecursive filter. The enve-
lope is observed to contain a series of N maxima
which have random amplitudes and occurrence times.
We assume that the occurrence of a maximum, or
peak, in the reflected signal envelope indicates
the presence of an irregular reflector. This is
the same assumption used in the equipment produc-
ing sonograms in the clinical environment. The
time series is divided into N nonoverl appi ng data
segments corresponding to the signals returned
from the individual reflectors.
The shape of the reflected signal envelope sug-
gests a way of segmenting the reflected data. For
our reflected signal analysis the data was segment-
ed about the peaks in the envelope using the loca-
tions of the adjacent minima to delimit the seg-
ment. Each segment is then assumed to represent
the return from an irregular reflector, with the
position of the envelope peak in the total re-
flected signal defined as the reflector location
within the tissue. This procedure produces a set
noi se
P(4>i)
4.34
I exp
'I'i
exp
8.68
N^N(0,ai)
For the emitted
reflected signal
(b)
Fig. 2. Reflection experimental configuration and
procedure. An acoustic pulse is emitted
and the reflection from the liver tissue
parenchyma are detected,
signal shown, a typical
contains a series of peaks which have ran-
dom amplitudes and occurrence times. These
are most evident in the envelope. The
received signal is divided into segments
delimited by the envelope mimma. The
power spectra are calculated and e is
estimated from the spectral differences
using least squares.
of nonoverlapping data segments which are assumed
to be statistically independent.
The data segment corresponding to the k^h re-
flector will be denoted by yk(n), n = 1,2, . . . jM^Ts-
The sample power spectrum, denoted by S|^(f), is
calculated from y|<(n) by using the squares of the
cosine and sine transforms given by
T l/'Vs
.(f) = — 11 E y(n)cos2TTfn
k > \n=l
(10)
X] y(n)sin 2iTfn
n=l
127
Because of the symmetry of the power spectrum it
suffices to consider the positive frequencies only.
In a manner similar to the transmission experi-
ment we will calculate the differences between the
sample spectra to estimate the value for b. For N
segments there are N/2 independent spectral dif-
ferences as shown in figure 3. From each observed
spectral difference we will estimate a value for 6,
denoted by 6k foi" the kth spectral difference.
The estimate for 6 using the entire reflected sig-
nal will be taken as the average of these individ-
ual estimates. A technique for pairing the seg-
ments to minimize the estimator variance will be
discussed later.
The power spectrum of the incident pulse has
significant energy only in a band of frequencies,
from fL (for lower) to fy ( for upper). It is in
this usable range that the log power spectral dif-
ference given in eq. (1) will be approximately
linear. Outside this range noise effects cause
random deviations. The set of frequencies f^
separated by the fundamental frequency spacing
are confined to be within this usable range, i.e..
(11)
LIVER
Emitted signal
Reflected
signal
Envelope
Segments
Set of
log spectral
di fferences
(b)
Fig. 3. Probability density functions. With the
rough boundaries modeled as linear filters
with random zero-mean Gaussian impulse
responses, the values of the power spec-
trum at the harmonic frequencies are
independent and distributed as the func-
tion shown in (a). The values of the
power spectral difference at the harmonic
frequencies are also independent and dis-
tributed as the function shown in (b).
In the analysis we shall assume that there are m
frequencies within the usable range, with fj de-
fined as the smallest.
5. Distribution of the Log
Spectral Difference
Now let us consider the log spectral difference
between the returns from two irregular reflectors,
i and j (a<j), separated by a distance d^j. If the
corresponding power spectra of the reflected signal
are denoted by S|^(f) and Sj(f), then their log spec-
tral difference in dB at the harmonic frequencies
is equal to
z. = 10 logS^(f.) - 10 logSj.(f.)
= 10 logs (f ) - 10 logSy (f.
+ n^(fi) - n (f.;
(12)
for i = 1 , 2, . . . , m
where ^X^ (f), k = n, j, is the power spectrum of
the propagating pulse at the k^h reflector. For a
tissue with a linearly increasing attenuation with
frequency, the difference between the propagating
pulse log spectra is the same as the transmission
experiment results given in eq. (1). In our case,
it is equal to
10 logS„ (f.)
for i = 1 , 2 ,
10 logS^ (f^)
(13)
If the difference of the spectral distortions at
the harmonic frequencies is defined as
^i = "£(^)
n.(f.) = n. „ - n. .
f (MHz) then eq. (12) can be written as
z. = 23d^jf . . s.
for i = 1 , 2,
(14)
(15)
In words, this last result states that the log
spectral difference between two reflected signals
at the harmonic frequencies is a distorted linear
function of frequency with a slope equal to the
value of 6, the coefficient to be estimated, and
the round-trip distance between the reflectors.
128
The distortion enters as an additive noise term.
Since e-j is the difference of random variables
which are independent and identically distributed
(iid) at the harmonic frequencies, ei is also iid
and the variances are equal at the harmonic fre-
quencies. Therefore, we can2drop the i subscript
a^d denote the variance by a^. We can show that
oe is independent of the individual reflector vari-
ances and is equal to 62 (dB)^ [7]. Therefore, the
distortions to the straight line calculated by
using the log spectra from any two reflectors will
have the same variance as the reflectors are spaced
farther apart (larger d^j) the slope of the log
spectral difference will increase. But since the
statistical roughness is assumed to be constant for
all reflectors, the distortions should remain the
same. We shall use this fact later in how we pair
segments to calculate the spectral difference.
Since the attenuation coefficient g affects only
the slope in eq. (15), the mean value of the dis-
tortion e-j can be eliminated by translating the
axes. This can be ac^compl i shed by destining new
variables Z-j = z-j - z and F.j = f-j - f where
and
i = l
(16a)
(16b)
The statistical model with zero mean distortion
and centered frequency axis can be written as
N' differences from which to estimate 6, where
N' 2 N/2. The statistical model using the entire
reflected signal can be written
Z., = 23d, F, +
^•k
(20)
for i
k
1, 2,
1, 2,
. , m
. , N' <
where Zj|^ is the observed value of the k^h log
spectral difference (after the mean value is sub-
tracted) at frequency F-j using signals from re-
flectors spaced d)^ apart.
6. Derivation of Maximum Likelihood
Estimator for 6
The results of the previous section indicated
that the additive distortions to the log spectral
difference are closely approximated by independent
and identically distributed Gaussian random vari-
ables with mean zero and variance a^.
Under these conditions the maximum likelihood
estimate of g, denoted by BmL' is found to be
equal to
N'
k=l
k^k
^ML
N'
k=l
(21)
where
for i = 1 , 2,
m.
(17)
The pdf of the additive noise term with zero
mean can be shown [7] to be equal to
1
1
4.34
*i
+ e
(18)
1 i=l ^
(22)
2d
i = l
is the maximum likelihood estimate of g using only
the information in the kth spectral difference.
It can be shown that this estimator is efficient,
i.e., it has the smallest variance of any unbiased
estimator [7]. The variance of the resulting
estimator can be shown to be equal to
We plot p(4>-j) in figure 2b. For comparison we
also s^ow the Gaussian pdf having mean 0 and vari-
ance Oj. = 62, in dashed lines. Because of their
similarity, we will approximate ifi-j by a Gaussian
random variable having mean 0 and variance a^, i.e..
i.e..
Var
hi] =
k=l
m
z
i = l
(23)
For notational convenience let us define S as
1 ^"^l
P(<f'-i) = — e ^
^ V2tt a
e
(19)
S =
i = l
(24)
The statistical model given by eq. (17) de-
scribes the log spectral difference between 2 re-
flectors separated by distance di^. The entire
reflected signal contains H segments from which we
can calculate N/2 independent log spectral differ-
ences. We will show later that depending on how
the segments are paired it may be desirable to use
The estimator variance given in eq.
be written as
(21) can then
Var
hi] -
N'
4r d2
k=l '
(25)
129
We want to minimize the estimator variance by
properly selecting the segments used in calculating
the spectral difference. If adjacent peaks are
chosen, all the du's will be small and the variance
will be large. Physically, this occurs because the
acoustic attenuation doesn't affect the signal
significantly for small tissue distances while the
irregular boundary effects are large. If we take
segments far apart so that the effect of acoustic
attenuation is more apparent, the number of spec-
tral differences N' becomes smaller. We will find
the optimum separation by considering a simple
tissue model .
For heuristic purposes let us consider a statis-
tically homogeneous tissue L cm long and having a
reflector density equal to y reflectors per cm.
The average spacing between reflectors is 1/p and
the average number of peaks in the reflected signal
is N = yL. If the segments to be used in calculat-
ing the log spectral difference are chosen such
that the distances between them are constant and
equal to d^ = n/y, N/2 5 n f N-1, the number of in-
dependent differences that can be formed is equal
to N' = N-n. The minimum variance can be shown to
occur for n = 2N/3. Then from eq. (25) we find the
minimum variance to be
Var
16
27 ^
(26)
The ED denotes equal d^istance separations between
reflectors. To relate this to tissue size we sub-
stitute N = yL to get
VarFLj .
L EDJmin
27
(27)
Recalling that yL = N we can write eq. (27) as
Varje^J . =
L EDJmin
— N.| 2
27 ^
(28)
This shows that in addition to decreasing as 1/N,
the variance also decreases as l/L^. This demon-
strates quantitatively the intuitive result that
for larger tissues, the acoustic attenuation be-
comes increasingly discernible relative to the
boundary distortion. ^
7. Confidence Intervals for the Estimator
In the previous section we derived the vari-
ance of the ED estimator as a function of tissue
size and reflector density. For Gaussian random
variables, the 95 percent confidence interval is
defined as an interval 1.96 times the square root
of the variance about the estimate of the param-
eter. From eq. (27) the confidence interval
width is given by
\ „1/2l3/2
(29)
with Cj = /27/1 6 S. For a given reflector density
y the confidence intervals will indicate the tis-
sue size required for a desired estimator resolu-
tion.
We will now present some experimental results
using refrigerated and formalin-fixed liver tis-
sue specimens.
8. Experimental Results
A 2.0 MHz transducer was used to obtain data
from in vitro refrigerated and fixed liver speci-
mens. The thickness of the refrigerated liver was
6.7 cm. After 1 month of formalin-fixation the
tissue size was 7.5 cm. From transmission tests
the average values of 6 were observed to be equal
to 0.63 dB/cm-MHz for refrigerated liver and 1.20
dB/cm-MHz for formalin-fixed liver. These values
agree favorably with those observed for refriger-
ated liver by Pauly and Schwan [1] and for formalin-
fixed liver by Lele [2]. The usable bandwidth was
determined from the transmission experiments to be
1.2 to 2.4 MHz. The envelope of the reflected
signal was calculated by using a 15 point lowpass
filter with a Gaussian shape truncated at ± 3 a.
This particular filter was chosen because its im-
pulse response matched the envelope of the incident
pulse. With this filter the average density of
peaks, y, in the reflected signal envelope was
observed to be equal to 7. The data were parti-
tioned into non-overlapping segments delimited by
the locations of the envelope minima. The same
power spectrum was calculated for each segment by
using eq. (10). The coefficient s was estimated
from the difference of spectra separated by a dis-
tance approximately equal to 2L/3, where L is the
total size of the tissue.
In the above analysis we were concerned with
the estimator performance as a function of tissue
size. In order to verify the predicted perfor-
mance of the confidence intervals with respect to
tissue size, small tissue sizes were simulated by
taking the equivalent amount of data from the total
signal reflected from the liver, starting at the
beginning of the record. The results then indi-
cate the experimental performance of the estimator
as the tissue size is increased. The confidence
intervals, instead of being taken about the esti-
mated value, for purposes of comparing the results
of different runs, are taken about the average
values of B observed from the transmission ex-
periments.
With the 15 point filter the average segment
size was 18 samples at Tj = 0.1 ys, producing a
fundamental frequency equal to 0.6 MHz. Two in-
dependent values per spectrum, located at 1.5 (Fi =
-.3) and 2.1 (F2 = +.3) MHz were used. From eq.
(24) the resulting value for S is 345. For y = 7
the variance given by eq. (Z7) is equal to
Var
K] -
(30)
From eq. (29) the resulting 95 percent confidence
intervals are given by
CI
18
18
(31)
The ED estimates as a function of tissue size
for 4 independent (nonoverlapping) runs, denoted
130
by Rl through R4, through refrigerated liver are
shown in figure 4. The 95 percent conficence in-
tervals are indicated. From the figure it is noted
that the confidence intervals are a reasonable pre-
diction of the estimator performance: wider vari-
ations and extreme values are found at the smaller
tissue sizes with the variations decreasing as the
tissue size increases. The confidence intervals
are rather wide indicating that large tissue sizes
are required to determine e accurately with this
technique.
The estimates for 6 independent runs, Fl to F6,
through fixed liver are shown in figure 5. The
confidence intervals are centered about the average
value of 6 observed for fixed liver. Again the
confidence intervals reasonably predict the observ-
ed performance.
The averages of the independent runs for refrig-
erated and fixed livers are shown in figure 6. The
confidence intervals have been reduced to account
for the averages of independent values (by 1/2 for
refrigerated liver, by 1//6 for fixed liver). Here
again the confidence intervals are consistent with
the observed results. Even for the averaged esti-
mates the reduced confidence intervals are still
B (dB/cm-MHz)
2.8t \ ■
REFRIGERATED LIVER
ESTIMATES
e (dB/cm-MHz)
FIXED LIVER ESTIMATES
-1.2
Fig. 4.
B estimates for refrigerated liver. The
theoretically derived 95 percent confidence
intervals are shown in dashed lines as a
function of liver size. The curves are
centered about the value observed from
transmission results. The larger the liver
the more data is produced and the better
the resulting estimate. Experimental re-
sults from four runs Rl to R4 are shown.
Smaller liver sizes were simulated by tak-
ing equivalent sections of data from the
total reflected signal.
estimates for formalin-fixed liver. The
confidence intervals are shown in dashed
lines as a function of liver size. The
curves are centered about the value ob-
served from transmission results. Experi-
mental results are shown for six runs
Fl to F6.
overlapped at 8 cm. Even though the experimental
results shown here differ noticeably for refrig-
erated and fixed liver, the fixed liver estimates
are close to the upper confidence interval predict-
ed for refrigerated liver.
9. Conclusion
The first order model of the irregular reflector
presented in this paper offers a reasonable de-
scription of the actual results obtained from re-
frigerated and fixed liver samples. The resulting
confidence intervals, however, are wide and overlap
significantly indicating that a large amount of
data is required for a reasonable estimator resolu-
tion. The spread in the estimates predicted by the
confidence intervals are due to the wide variations
possible in the reflected signal power spectrum
caused by the white Gaussian impulse response model
of the irregular reflector. To decrease the con-
fidence intervals, and hopefully the resulting ex-
perimental estimates, a more accurate model of the
typical reflector is required. In the reflector
model assumed here we have ignored possible de-
terministic reflections and have assumed the re-
flector effects to be completely random. We plan
to study an improved model, combining both de-
terministic and random effects, for which better
estimation techniques can be derived to produce
more accurate estimates from smaller tissue samples.
131
(dB/cm-MHz)
2.4
2.0
1.6
1.2
0.4
\ \
\ \
Averaged Estimates
>v
\ \
\\
V 0 Refrigerated
\ liver
\^ * Formal in-fixed
tv. \^ liver
\\ ^^~^^^95%-F
FORMAl IN-FIXED
\,^%-R"
_REFRIGERA1ED
1 — 1 1 1 1 1 —
1 2 ^3
-
^ — f—{ \ — 1^ \ 1 \ — 1
/4 / 5 6 7 (cm
/ /
/ /
//
1/
//
//
-1.2
Fig. 6. Comparison averaged B estimates for refrig-
erated and fixed livers. The respective
confidence intervals are indicated about
the values observed in transmission experi-
ments. The averages of the 4 refrigerated
liver runs and 6 formalin-fixed liver runs
indicate a separation in their values.
Acknowledgment
This research was made possible by the loan of
the fast analog-to-digital converter and paper tape
unit by the Picker Corporation.
References
[1] Pauly, H. and Schwan, H. P., Mechanism of
absorption of ultrasound in liver tissue,
J. Acoust. Soc. Am. 50, 792-99 (1971).
[2] Lele, P. D., Mansfield, A. B., Murphy, A. I.,
Namery, J., and Senepati , N. , Tissue Charac-
terization by Ultrasonic Frequency-dependent
Attenuation and Scattering, in Ultrasonic
Tissue Characterization, M. Linzer, ed.,
pp. 167-96, National Bureau of Standards Spec.
Publ. 453 (U.S. Government Printing Office,
Washington, D.C., 1976).
[3] Miller, J. G. , Yuhas, D. E., Mimbs, J. W.,
Dierker, S. B., Busse, L. J., Laterra, J. J.,
Weiss, A. N., and Sobel , B. E. , Ultrasonic
Tissue Characterization, 1976 IEEE Ultrasonics
Symposium Proc. , 76 CH 1120-5SU, 33-43 (1976).
[4] Kuc, R. , Schwartz, M. , and Von Micsky, L.,
Parametric Estimation of the Acoustic At-
tenuation Coefficient Slope for Soft Tissue,
1976 IEEE Ultrasonics Symposium Proc. , 76 CH
1120-5SU, 44-47 (1976).
[5] Elias, H. and Sherrick, J. S., Morphology of
the Liver (Academic Press, New York, 1969).
[6] Jenkins, G. N. and Watts, D. G., Spectral
Analysis and Its Applications (Holden-Day,
San Franci sco, 1968) .
[7] Kuc, R., Statistical Estimation of the Acous-
tic Attenuation Slope for Liver Tissue, Ph.D.
Dissertation, Columbia University (1977).
[8] Schwartz, M. , Information Transmission, Modu-
lation, and Noise (McGraw-Hill, New York,
1970).
[9] E. Parzen, Modern Probability Theory and Its
Applications (Wiley, New York, 1967).
132
CHAPTER 5
SCATTERING
133
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed.. National Bureau
of Standards, Spec. Publ . 525 (U.S. Government Printing Office, Washington, D.C., 1979)
AN ULTRASONIC TISSUE SIGNATURE FOR THE LUNG SURFACE
Theodore L. Rhynei
Massachusetts Institute of Technology
and Massachusetts General Hospital
Cambridge, Massachusetts 02139, U.S.A.
The normal lung surface is modeled as a statistically rough surface composed of
the air-containing alveolar sacs at the periphery. The ultrasonic reflection (scatter)
from the rough surface is analyzed as a random process and a frequency signature, which
depends upon alveolar sac radii statistics, is predicted for the reflectance. Pulse-
echo ultrasonic instrumentation for the measurement of the absolute reflectance from
the lung surface is presented. Lung reflectance data confirming the lung surface fre-
quency signature is presented for humans. Applications of the model are discussed to
the detection of pulmonary embolism, atelectasis, pulmonary edema, and chronic lung
disease.
Keywords: Detection of lung disease; frequency signature; statistical scattering;
tissue signature; ultrasonic scattering; ultrasound.
1. Introduction
Ultrasonic characterization of tissue rep-
resents a considerable challenge to the re-
searcher in that he must exploit essentially
"second order" phenomena while current clinical
techniques have already exploited the "first
order" phenomena. The "second order" phenomena
(dispersion, reflection, scattering, movement,
etc. ) are often measurable only after suitable
instrumentation has been developed. The re-
searcher is especially challenged to select from
an infinity of possible instrumentation approach-
es and plausible phenomena in order to arrive at
a clinically useful diagnostic technique. The
work reported here is the development of a tissue
characterization for the normal lung surface using
an analytical approach. The approach is quite
possibly as significant as the results since it
offers a path through the complex decisions in-
volving instrumentation design, study of phenomena,
and definition of the useful measurement.
The analytical approach used here is summarized
in the four interactive steps outlined below. The
interplay of each step can be followed as the
normal lung model is developed.
1) Calibration of the transmitter-receiver
circuitry and transduction process to either an
absolute or traceable standard for measurement of
tissue acoustic properties that are independent of
the instrumentation.
2) Design of the transmitted acoustic signal
waveform to accentuate the desired tissue phe-
nomena.
3) Development and validation of analytical
models for the tissue target which characterize
^Current Address: Bolt, Beranek and Newman, Inc.,
50 Moulton Street, Cambridge, MA 02138.
the measurable acoustic phenomena of tissue in the
normal and diseased states.
4) Development of signal processing operations
performed by the instrumentation to detect and
quantify echo parameters that characterize the
normal and diseased states of the tissue.
2. Normal Lung
The lung is a very large organ whose surface
lies within a few centimeters of the skin over
most of the chest. As shown in figure la, a trans-
ducer coupled to the skin may direct ultrasonic
waves through the soft tissues of the intercostal
space that lies between adjacent ribs. If a suffi-
ciently small transducer is utilized, the lung
surface can be examined with no interference from
nearby ribs.
A normal lung has a gross density that is ap-
proximately one third that of water. Thus, the
lung is approximately two thirds air and one third
soft tissue. At the periphery the schematic of
the magnified lung surface in figure lb indicates
the relationship of the air-containing alveolar
sacs to the soft tissues that surround the sacs.
The alveolar sac is the end-respiratory airway and
is divided up into the familiar alveolar chambers.
The marked difference in acoustic impedance be-
tween the air and the surrounding soft tissue
causes nearly perfect reflection at the air to soft
tissue boundary. For our purposes we postulate an
"ultrasonic" lung surface which is a roughened
planar surface composed of the outer walls of the
multitude of peripheral alveolar sacs that lie just
inside the pleural membrane.
The physical lung surface (as opposed to the
"ultrasonic" surface) is a pleural membrane which
forms a unique bearing with a second pleural mem-
brane that is attached to the inside of the chest
135
Lung Surface (Pleura)
Skin and Soft Tissue
(a )
Pleural Surface
Alveolar Sac
(Diameter «275;i.m)
(b)
Fig. 1. Schematic representation of the chest
and lung surface; (a) cross section of
the chest wall; (b) enlarged cross sec-
tion of the lung surface.
wall. During respiration the lung moves lateral-
ly within the chest on this pleura-fluid-pleura
bearing. Since the bearing is composed of soft
tissue and is only a few tens of a micrometer
thick, it does not contribute a significant echo
in comparison to the "ultrasonic" lung surface.
Furthermore, the lateral motion of the lung per-
mits the lung surface to be scanned during respira-
tion by a transducer held stationary with respect
to the chest wal 1 .
The geometry of the lung examination is provid-
ed in figure 2. A radiating and receiving trans-
Transducer Aperture Rough Surface of
Infinite Extent
Fig. 2. Geometry for the examination of the lung
surface by a transducer with a circular
aperture.
ducer located at the skin is represented by a circu-
lar disk which is the aperture of the transducer.
The radiating and receiving disk observes the lung
surface through the soft tissue of the intercostal
space, which is a uniformly lossy media. The
"ultrasonic" lung surface is a rough surface normal
to the disk and of infinite extent. The rough sur-
face may move laterally with no change in range,
thus presenting new portions of its surface to the
area directly underlying the disk.
3. Calibrated Instrumentation
and Signal and Design
Wave propagation between the radiating (and re-
ceiving) aperture and lung surface forms a central
issue in the calibrated measurement of echoes re-
flected from the lung surface. If we further
simplify the situation represented in figure 2 to
a lossless media and a flat perfectly reflective
plane, we have a classic radiation (or diffraction)
problem. By virtue of the finite size of the
radiating and receiving disk plus the wave speed
of the media, waves launched by the disk produce
filtered echoes as they reflect from the plane and
return to the disk. The radiation (or diffraction)
filter response is given in Appendix A as developed
in the literature [1]^. If the response is graphed
as in figure 3 for a representative situation, we
see that the radiation filter is a high pass filter.
The nearly flat response of the radiation filter
depends on the selection of aperture size, fre-
quency, and range so that the surface lies within
the near-field of the transducer. This is the
case for the lung surface measurements. Sub-
sequently, the radiation filter will be used both
for calibration and adjustment of lung surface
data.
cr ■
-5 - .
-6 1— — ^ ^ 1 1 1 I I I I
0 I 23456789 10
Frequency MHz
Fig. 3. Frequency response of the radiation
filter between a radiating disk of
radius 0.0125 inch and a flat perfectly
reflecting plane 1.5 cm distant.
The pulse echo instrumentation is diagrammed in
the network of figure 4. It consists of a trans-
mitter, a receiver, and a transducer, each linear- '<
ly coupled to a common transmission system. The
transmitting waveform excites the system. The
waveform V^(t) propagates through the transmission
line to the transducer. The transducer couples
the electrical domain to the acoustic radiation
field. Energy launched by the radiating disk of
figure 2 is represented as energy consumed in the
radiation resistance. Conversely, echoed waves
striking the disk are represented by the wave '
^Figures in brackets indicate literature
references at the end of this paper.
136
Radiation
Impedance
+
Wave
Potential
Transducer
Loss
Lossless
Lossless
Matching
Transducer
Network
I Transmission Line
Transformer of Impedance Rg
Receiver
Outputs
Fig. 4. Schematic of pulse-echo ultrasonic instrumentation.
potential Vg(t). The radiated waveform is Vt(t)
as filtered by the transmission line and the
transducer. The received waveform Vf-(t) is the
wave potential Vg(t) as filtered once again by the
transducer and transmission line.
If the transmitter (plus receiver) matches the
impedance of the transmission line, then the filtra-
tion of the transmission line consists of simple
time delays of the transmitted and received sig-
nals. Also, the impedance presented to the trans-
ducer is Rq in series with V-t-(t). Thus, matching
effectively removes the transmission line from the
network of figure 4. The net effect of the in-
strumentation is the filtration of transmitted and
received signals by the transducer.
In order to measure transducer response and the
magnitude of echoes from a target, it becomes im-
portant to define a measure of absolute signal
loss. This is done by energy conservation of the
energy in the transmitted wave V^(t), a portion of
which ultimately returns as the received echo
Vp(t). Loss is defined here as the ratio of the
exchangeable power of the transmitter to the re-
ceived power in eq. (1). This is more convenient-
ly expressed as the gain (always less than unity)
or transfer gain in eq. (2) which can be expressed
in decibel or magnitude notation.
(V2)/(4R ) V.
Loss (dB) = 10 logio — 20 logiQ— (1)
(Vp/(R ) 2V
2V
gain = — ^ , (2)
Since the transmitted acoustic waveform is de-
termined by V|.(t), we are free to choose it so as
to accentuate some property of the target. Here,
the magnitude of the lung surface echo is the im-
portant property. Furthermore, the returned lung
echo will be corrected by the radiation filter
and transducer filter magnitudes. Therefore, we
select a transmitted waveform which is a carrier
burst of sufficient duration to allow the trans-
ducer and the radiation process to reach their
steady-state or Fourier responses. The trans-
mitted pulse waveform is completely described by
the magnitude, frequency, and duration of the
sinusoidal carrier. Conversely, the carrier
burst must be sufficiently short in time to allow
range discrimination of the lung echo from other
echoes. The values for and utilized in loss
calculations are the instantaneous values of the
envelopes of carrier waves.
The transducer frequency response is experi-
mentally determined as an absolute quantity using
the selected signals, the definition of loss, the
radiation filter correction (fig. 3), and an ex-
perimental geometry as in figure 2 (with a flat
perfectly reflective planar test target) [2]. It
is interesting to note that absolute dosimetry may
be determined as the power dissipated in the radia-
tion resistance of figure 4 [2,3]. The transducer
of figure 4 provides a transformer and a lossless
matching network in addition to the transducing
plate and associated radiation and dissipation
loads. Modeling of the transducing disk is ex-
tensively discussed elsewhere [2,3] where it is
shown that a simple model for transducer gain
exists at half-wave resonance. Also, it is shown
that optimum loop sensitivity results from adjust-
ment of the transformer for impedance matching of
transmitter to transducer impedances. The adop-
tion of a transmission system impedance plus
specification of transducer transfer function
could form the basis of standardization for pulse
echo instrumentation.
4. Scattering Model for the Lung Surface
Returning to the measurement of the reflection
from the lung surface depicted in figure 2, we are
motivated by the microscopic structure to model
the lung as a randomly rough surface. The essen-
tial difference between the flat surface of figure
3 and a rough surface is that the rough surface
has a finite thickness or scattering depth and is
said to be range-spread. Mathematically, this is
described in eq. (3) as an integration over all
range of some range-spread function m(') with the
impulse response of the radiation coupling
f(',-,-)^ which is the time domain pair to eq.
(A-1).
00
r(t) = f f(r,a,t) m(r) dr (3)
For a "thinly rough" surface we may expand the
function f(r,a,t) about some fixed range (rp),
make a change of variables for fixed wave speed c,
and thus turn the equation into the time convolu-
tion in eq. (4).
137
r(t)
J f(r^,a,t-T)
c_ m(CT)
2 2
dx
(4)
Equation (4) is highly significant in that the re-
ceived signal r(t) is the result of a linear filtra-
tion in cascade with the radiation filter of eq.
(A-1). The frequency equivalent of eq. (4) is
simply a product of transfer functions in eq. (5),
where G(f) is the transfer function associated
with m( • ) •
R(f) = F(r,a,f) G(f)
(5)
The filter G(f) and its transform pair he
m(%CT) represent the filtration of a randomly
rough surface and, therefore, are themselves ran-
dom. In a sense the precise G(f) depends upon
the particular ensemble of scattering elements
that lie in the field pattern of the transducing
disk of figure 2. The value of the Fourier func-
tion G(f) is the magnitude (and phase) of the re-
flection due to a long carrier burst. By allow-
ing the transducer, radiation filter, and reflec-
tion filter to reach steady-state as discussed
above, one may make direct measurement of the
magnitude of G(f). The measured value will vary
as different ensembles of scattering elements
sweep in front of the transducer. The charac-
terization of the lung surface lies in relating
the statistical properties of the lung surface to
the probability density function (and its mean and
variance) for the random variable |G(f)|.
A randomly rough surface representing the normal
lung consists of a planar array of the alveolar
sac wall segments shown in figure 1(b). Between
adjacent alveolar sac chambers cusp-like regions
are formed which contain soft tissue which readily
support wave propagation. The depth of these
cusps constitutes the thickness or scattering
depth of the "ultrasonic" lung surface. The lung
surface is characterized as having a mean scatter-
ing depth (wq) and standard deviation of scatter-
ing depth [o^). Figure 1(b) indicates that the
scattering depth Wq is approximately equal to the
mean radius of the peripheral alveolar sacs. If
the wavelength of the ultrasound is very much
greater than Wq one expects the surface to reflect
as though it were flat. Conversely, at some criti-
cal wavelength related to the mean scattering
depth Wq one would expect destructive interference.
The degree of the destructive interference would
of course depend upon the regularity of the scat-
tering surface with greater destructive inter-
ference occurring for smaller a^.
Using the above model for the lung surface plus
several additional assumptions the probability
density function plus its mean and standard devia-
tion for |G(f)| are predicted in Appendix B. The
density function is a Ricean function whose shape
depends upon Wq, a^, and the frequency f. The
most significar[t measurable properties of |G(f)|
are the mean (Z) and standard deviation (a^) given
in figure 5 as functions of frequency. At lower
frequencies the curves approach those of a perfect
reflector (OdB). Conversely, there is a destruc-
tive interference minimum at a frequency where
one-half a wavelength equals the mean scattering
depth Wq. Figure 5(a) demonstrates the dependence
of the dip frequency on Wq. In figure 5(b) we
note that the depth of the dip increases for great-
er regularity in the scattering depth (smaller c^) .
?-5
■^-10
o
tr
1-20
o
^-25
qI
-30
1 1 1 1 1 \ 1 1 r
Mean Z
leO^nn
Standard Deviation o-.
J_
4 5
Frequency MHz
(a)
4 5
Frequency MHz
(b)
Fig. 5. Families of curves for the mean Z and
standard deviation of the predicted
echo amplitude, dB versus frequency, (a)
curves for two values of scattering
depth Wq and fixed a^; (b) curves for
two values of a^^ and fixed Wq.
5. Experimental Results
Experimental measurement of the lung surface
reflection can be made to confirm the above model.
Measured values are corrected for radiation,
transducer, transmitted voltage, and bulk tissue
effects as in eq. (6). The resulting experimental
value (Y) is
,(l/20)2rf0.897 IQ-^
2V 10'
r
|F(r,a,f)| |T(f)P
(6)
is a measure of the absolute reflectivity |G(f)| of
the lung surface at the frequency f. In the equa-
tion Vr and Vt are envelopes of sinusoids as dis-
cussed earlier. The function |T(f)|2 is the mea-
sured transducer loop transfer function and
F(r,a,f) is the radiation transfer function for
range r and frequency f. Correction is made for
bulk tissue loss using an attenuation constant of
0.897 dB/cm.
By averaging over many values of Y at a single
frequency for successive pulse echo transmissions
with the lung surface scanning before the_trans-
ducer, we arrive at the mean reflectance Y for
138
m -5
o- 10 -
=£-15-
-20
g-25
-30
Mean Y
Standard Deviation a„
4 5 6-
Frequency MHz
Fig. 6. Experimental data for the mean and
standard deviation of lung reflection Y
for a normal human volunteer.
that frequency. Similarly, one may compute the
standard deviation (a^) for Y over the data for
a single frequency, when multiple frequencies
are examined and the data reduced for Y and oy,
the experimental data in figure 6 is arrived at.
Note that the destructive dip is present near
5.2 MHz and the curve rises toward 0 dB at lower
frequencies. A more extensive discussion shows
close correlation between the predicted proba-
bility density function of Appendix B and ex-
perimental histograms [2,31. These data confirm
the analytical model of tissue thus completing
step 3 in the approach to tissue characterization.
6. Applications
The final element in the approach to tissue
characterization is to define the signal process-
ing functions of figure 4 to perform some clini-
cally useful application. Two areas of applica-
tion are suggested for the normal lung tissue sig-
nature developed here, they are: a) detection of
acute infiltrative disease and b) detection of
chronic pulmonary disease.
The normal lung surface strongly reflects the
ultrasound due primarily to the presence of air
at the periphery. Several diseases, including
pneumonia, edema, atelectasis, and pulmonary em-
bolism, which acutely alter the air content at the
periphery should be detectable as a departure from
the normal model. Preliminary clinical evidence
suggests that reductions in lung reflectivity from
10 dB to 30 dB are associated with these acute dis-
eases [2,3]. An optimal method of detecting the
reduction in reflection has been suggested [2,3]
utilizing the predicted probability density func-
tion to determine a sufficient statistic which is
the root mean square reflection. An ultrasonic
instrument embodying this technique is shown in
figure 7. The performance of this detection
scheme can be evaluated in terms of probability
for false positive and false negative with the
testing threshold (T) set to optimize performance.
The chronically abnormal lung can be postulated
to produce reflection functions differing from
those of figures 5 and 6 due to rearrangement of
tissue structures at the periphery of the lung.
Diseases such as emphysema and pleural diseases
are of clinical intere^st. Figure 8 demonstrates
the reflection data (Y,ay) for a subject with
severe chronic pulmonary disease. The destructive
interference dip has been suppressed presumably by
the higher variance in the scattering depth (a^)
of the diseased structures at the periphery. In-
deed, if a greater value of a^j is used in the model
of figure 5(b) the dip becomes obliterated. A sug-
gested application to chronic disease is to fit
predicted curves to experimental data and take the
resulting scattering parameters Wq and as in-
dices of the lung surface [2,3].
The above applications depend upon detecting
departures from the normal model. Further develop-
ment of this tissue characterization approach sug-
gests developing detailed models (step 3) for the
diseased states so that further clinical applica-
tions can be defined (step 4).
Transducer
Lung Surface
Transnnitter
•lF(r,a,f)l Vt lT(f)|2-
.7;
Meter
Light if
Less Than
Threshold
RMS Envelope
Comparator
2 ll/2
Sample Lung Echo
T Threshold
Fig. 7. Ultrasonic instrument for detecting acute lung disease using
a sufficient statistic for testing against a threshold.
139
CD
"°-5
;-io
-15
^-25 -
-30
"I 1 1 1 1 1 r
Mean Y
2^-20- -_ -
Standard Deviation ,
J I 1 L_
J L
4 5 6
Frequency MHz
Fig. 8. Experimental data for the mean and
standard deviation of lung reflection Y
for a subject with severe chronic pulmo-
nary disease.
Appendix A
Radiation Filter Response
The radiation filter response reported in the
literature [1] and tabulated [2] is given in eq.
(A-1) for the definitions in eqs. (A-2). It is an
exact solution to the radiation coupling between
two coaxial disks spaced at distance 2r apart.
F(r,a,f) = [cos(2Trft3) + j sin(2TTft3 )]
(A-1
c^
- t3ti,
;.2
"tu + t,
Lt, + ti
jQ(2^ft3) + jJi(2TTft3
rtu + t,
+ — tf
In the equations c is the wave speed, f is the
frequency, a is the disl< radius, and r is the
range between disl< and plane. The functions Jq,
are Bessel functions of the first l<ind.
Appendix B
Detailed Scattering Function
for the Lung Surface
This appendix is a summary of a more detailed
development of the scattering function for the
lung given elsewhere [2,3]. The range spread
function m(r) is considered to be the incremental
reflectance of the rough surface due to surface
elements between r and r+dr in range. In order to
simplify the development the following assumptions
are made:
1) Reflection is a first order scattering
process involving no shadowing of reflective ele-
ments nor multiple reflection among elements.
2) The dimensions of roughness elements are
small compared to the diameter of the transducer.
3) The soft tissue to air interfaces perfect-
ly reflect with no other loss mechanism.
4) The disl< is normal to the surface.
5) The scattering depth is thin compared to
variations in the radiation transfer function.
6) The incremental reflectance from an in-
dividual scattering element is of constant ampli-
tude over the depth of the element.
7) The scattering elements are statistically
independent.
8) The lung surface approaches a flat perfect-
ly reflective plane as the frequency approaches
zero.
The transfer function for an individual scat-
tering element of depth Wjv, is given in eq. (B-1).
The overall transfer function is a sum over a
large number of individual scattering elements in
eq. (B-2).
L(f ,wni)
j^TTf
-j2TTf2wni
1 - e
(B-1
1 . +
tu + tl t. + t;
J^(27Tft3;
Jl (27Tftc
+ higher terms
tl = 2r/c
[432 + 4r2
27Tft.
+ jJl(2TTft:
(A-2a)
(A-2b)
F(f) = h E L(f,wm)
M=l
(B-2)
Invoking the central limit theorem G(f) becomes a
Gaussian random variable with independent real and
imaginary parts with means R^, Qq and variance a^.
When the value of k in eq. (B-1) is adjusted to
normalize the magnitude as in assumption 8, the
essential values for Rp, Qq and are arrived at
in eqs. (B-3) and (8-4).
4w2(2Trf)2
1 + e
-4aw2(2Trf )2
r2
t3 = h{t2 - tl)
= h[t2 + tl)
(A-2c)
(A-2d)
. .^ -2aw2(2Trf)2
(2wo2uf) ^2—^
2 cos e
(B-3)
140
8w2(2TTf)2
-4aw2(2Trf )-
If we are concerned only with the magnitude Z
of G(f) the probability density function becomes
Ricean as in eq. (B-5). The mean and a variance
of the magnitude are given in eqs. (B-6) and (B-7)
respectively. The functions lo(') and lii-) are
modified Bessel functions.
References
^"4) [1] Rhyne, T. L., Radiation coupling of a disk
to a plane and back or a disk to a disk: An
exact solution, J. Acoust. Soc. Am. 61 , 318-
324 (1977).
-(Z2 + R2 + Q2)
.(Z) = ^ e
2a2
Z(R2 + Q2)
0 0
[2] Rhyne, T. L. , Acoustic Instrumentation and
Characterization of Lung Tissue (Forest
Grove, Oregon, 1977).
[3] Rhyne, T. L. , Sonar Characterization of Tis-
sue as Applied to the Lung, Sc. D. Thesis,
Department of Electrical Engineering and Com-
puter Science, Massachusetts Institute of
Technology, Cambridge, Massachusetts (1976).
Z > 0
(B-5)
-(R^ + Q^)
Z = e
4a2 a^Tt '
R2 + Q2\ R2 + Q2
1 + _o ^1 i_ 0 ^0
2t2
4t2
R2 + Q2 /r2 + q2
2t2
4t2
(B-6)
j2 =
z
Z2
-7)
The mean squared lung surface reflectance in
eq. (B-8) expresses the energy scattered at a
given frequency. This function is essentially
the scattering function discussed in radar.
|G(f)|
2 = ,2 =
R2 . Q2 . 2a2
(B-E
Acknowledgments
This paper is essentially a summary of work re-
ported in the three references cited. It was felt
that it would be useful to the reader to report
the essential results here in summary form and to
refer the interested reader to the more extensive
discussions and numerous literature citations to
be found in the basic papers.
The work summarized here and in the basic
papers was supported by the Ambrose Monell Founda-
tion and the G. Linger Vetlesen Foundation.
141
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed., National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
ANGLE SCAN AND FREQUENCY-SWEPT ULTRASONIC SCATTERING CHARACTERIZATION OF TISSUE
R. C. Waag, P. P. K. Lee, R. M. Lerner, L. P. Hunter, R. Gramiak, and E. A. Schenk
Departments of Electrical Engineering, Radiology, and Pathology
University of Rochester
Rochester, New York 14627, U.S.A.
Ultrasonic wave interference has been applied to characterize tissue by measuring
scattered wave intensity as a function of frequency and angle. A theory of acoustic
wave propagation and scattering in an inhomogeneous medium has been employed to show
that received ultrasound may be expressed in terms of medium refractive index varia-
tions via a Fourier transform. Using a computer-based system, measurements were made
on model targets and post-mortem liver specimens. Model studies demonstrate that
regular scatterer spacing can be inferred from measured diffraction data by Fourier
inversion and that scattering differences can be observed from random media consist-
ing of particles of different average sizes. Scattering from liver indicates there
is significantly more energy scattered at small angles than is backscattered. Cor-
relation of ultrasound scattering with structure observed through a microscope has
been obtained by computing the diffraction pattern of the two-dimensional optical
transmittance images acquired through a microscope-TV chain attached to the computer.
Average particle sizes of the model random media determined by Fourier analysis that
exhibited diffraction rings of the digitized cross-sections yielded scattering predic-
tions which were in agreement with measured acoustic data.
Key words: Angular scattering; characterization of tissue; optical correlation;
scattering; swept-f requency diffraction.
1. Introduction
Although ultrasound has been widely accepted
as a valuable diagnostic tool in medical imaging,
its use in quantitative characterization of tis-
sue is just beginning to emerge. Among methods
which employ amplitude, phase, frequency, and at-
tenuation of ultrasound signals, diffraction-
based techniques offer the potential of charac-
terizing acoustic scattering element distribu-
tion on a scale corresponding to the wavelength
within a finite volume probed by the ultrasound
beam [l-6]i. Success of wave interference methods
in x-ray studies of materials as well as in atmos-
pheric probing by radar motivates the development
of the methodology for acoustic studies of tissue.
However, realization of practical tissue charac-
terization systems based on diffraction require
more detailed understanding of the acoustic scat-
tering process, including development of useful
models and knowledge of practical measurement
limitations.
In this paper, our ultrasonic scattering studies
of model targets and preliminary results from liver
tissue are described. A model is developed to show
the fundamental relationship between scattered
acoustic waves and tissue structure. The measure-
ments demonstrate feasibility of the method and
identify problems. The results indicate the prom-
^Figures in brackets indicate literature
references at the end of this paper.
ise of the concept but point out the need for ad-
ditional data to characterize tissue.
Our research combining angle scanning and fre-
quency sweeping is intended to provide a founda-
tion for characterization of tissue by its ultra-
sonic scattering properties.
2. Theoretical Basis
A. Acoustic Scattering Model
The model we use relates the scattered intensi-
ty to the acoustic structure of the tissue and in-
dicates that different levels of organization may
be studied with ultrasound by appropriate choice
of frequency and geometry. The resulting acoustic
characterization can then be compared to micro-
scopic determinations of structure to establish
tissue pathology. The theory is analogous to that
used in x-ray diffraction of amorphous materials
and allows the structure of the medium to be in-
ferred from measurements of the scattered acoustic
wave.
The tissue is modeled as an inhomogeneous medium
in which the acoustic properties exhibit variations
from point to point. Since wave propagation is af-
fected by the local environment, wavefronts are dis-
torted and scattered. The derivation has followed
the basic approach used in modeling wave propaga-
tion and scattering for acoustic studies of the
lower atmosphere [7].
143
The analysis begins with consideration of a
plane wave propagating through a lossless unbounded
medium which has small variations in refractive
index. Small values of absorption do not change
the basic relationships [6].
Assuming a monochromatic incident plane wave,
weak single scattering, and a receiver in the far-
field, the resulting scattered wave velocity poten-
tial 't'llrjk) is given by
(t.i(r,k) =
A k2 e^^^ r ■
k-r'
2-nr
ni(r' )dV'
(1)
V
in which
Aq = incident plane wave amplitude
k = incident wave number
K = scattering wave vector
ni(r') = variations of acoustic refractive index
V = scattering volume
r = distance from the scattering volume
center to the reception point
Thus, the velocity potential at a receiving point r
is equal to the product of a frequency-dependent
factor and the three-dimensional spatial Fourier
transform of the acoustic refractive index varia-
tions.
The theory also relates the mean square value
of velocity potential to the correlation function
of the acoustic refractive index variations in a
random medium if the variations are stationary in
space. The relation is
:|4>i(r,k)|:
A2k'*V'
0
"1
(p)dV'
(2)
where
(p) = <ni(r' )ni(r' + p)>
where < > denotes ensemble averaging. This shows
that the average intensity of the scattered wave
is proportional to the three-dimensional spatial
Fourier transform of the correlation function of
the acoustic refractive index variations.
The expression for the mean square velocity
potential allows determination of the correlation
function of the acoustic refractive index varia-
tions in the medium by inversion of the transform
relating tbe measured average scattered intensity
as a function of scattering vector [8]. Alterna-
tively, a specific form for the correlation func-
tion can be employed to evaluate the average scat-
tered intensity for comparison with experimental
data. The latter approach has been used to evalu-
ate scattering from a cloud of scatterers assuming
a correlation function with a Gaussian shape modi-
fied to remove the dc spectral component [9].
Neglecting density variations, the resultant ex-
pression for the scattered intensity per unit in-
cident intensity is
(3)
k2a2sin2|)e
-2k2a^
.2^
V = scattering angle
NV = mean number of scatterers in sample volume
a = mean radius of the particles
= mean value of compressibility variations
between scatterers and medium
The result predicts that smaller particles scatter
more energy off-axis than large particles for a
fixed incident frequency.
B. Optical Correlation
Optical data for comparison to ultrasound angle
scattering data can be obtained by computing the
diffraction pattern of a sample region with vari-
able transmission representing spatial distribu-
tion of scattering inhomogeneities. In the far-
field of a plane through the region, the diffracted
optical field is given in terms of the two-dimen-
sional Fourier transform of the transmission pat-
tern [10].
Ae
-i kr
J ^^l >^ " -
dr'
(4)
where
A
k
T(r')
constant
2i r
\ r
transmittance of aperture S
The intensity of I(r) of the pattern is then
I(r) = *(r)**(r)
The dimensions over which diffraction occurs are
determined by the wavelength or, equivalently, the
sampling interval of the optical transmittance and
can be readily chosen in analysis to yield a scale
comparable to that in ultrasound diffraction.
Averaging the two-dimensional diffraction intensity
patterns of any collection of typical cross-sections
in the region results in a statistical characteri-
zation of diffraction by a random medium.
For an isotropic random medium with local order,
the predicted average diffracted intensity takes
the form of concentric rings [ill. The spacing of
the rings gives nearest neighbor distance while the
width of the rings is a measure of the spread in
scatterer size and the number of visible rings in-
dicates ranges of local order. Optical diffraction
also allows quantitative determinations of param-
eters describing known optical morphology for com-
parison with similar parameters derived ultrasoni-
cally when there is no order since the width or
general shape of the power spectra can be used in
these cases.
3. Experimental Methods
A. Acoustic Scattering
Initial studies employed frequency sweeping
with fixed transducer-target orientation. The
scattering experiments were performed in a water-
filled tank containing a fixture on which trans-
ducer holders were mounted. The holders could be
angled to align the ultrasonic beams and could be
slid along the fixture as well.
144
The electronic instrumentation used for this
investigation consisted of a transmitter capable
of frequency sweeping, a range-gated receiver and
a recorder. Bursts of ultrasound were gated from
a variable frequency master oscillator to a power
amplifier which drives a wideband 9.5 mm diameter
transducer with a 5 MHz center frequency. The
frequency of the master oscillator was slowly
swept over a range of 2 to 8 MHz by a linear ramp
function that produced negligible carrier frequen-
cy variation over any single pulse.
The scattered ultrasound pressure was detected
by a receiving transducer matched to that used for
transmission and the amplified RF signal was
energy-detected to give a video signal proportion-
al to the incident wave peak intensity which was
then recorded.
Arrays of cylinders were studied to provide in-
formation about the influence of geometric align-
ment and beam uniformity. The scatterer positions
for array measurements were established by five
sets of uniformly spaced grooves two centimeters
in length. A rubber sheet was used to shield the
holder exposing only one set of wires for acoustic
measurements. The center of the array was nominal-
ly set at the intersection of the transducer beams
for each study with the array elements lying along
the bisector of the angle formed by the beams.
The energy detector output was bandpass filter-
ed to produce a signal that was displayed as a
function of frequency on the face of an oscillo-
scope for photographic recording. The analog curv-
es were sampled at 0.115 MHz intervals and quan-
tized into integers in the range 0 to 255 for cal-
culation of Fourier transforms.
Ultrasound scattering as a function of angle
was measured in a water tank using an acoustic
diffractometer [12] that provided precise control
of transducer position relative to the scattering
volume. The scattering medium was contained in a
cylindrical column 28 mm in diameter by a .0254
mm thick cellulose tubing that minimized reflected
energy loss at the perimeter of the sample. Trans-
mitting and receiving transducers containing 4.76
mm radius ceramic disks as active elements were
rotated in equal increments but opposite directions
about the axis of the cylindrical sample. Trans-
mitter and receiver distances from the axis of
rotation were each equal to 13.6 cm. This scanning
procedure maintains a constant scattering vector
direction while changing the scattering angle, v,
and hence the scattering vector magnitude. The
centers of transducer rotation were maintained to
within ± 0.3 mm during the angle scan.
The electronics were the same as in the swept-
frequency measurements. The receiver electronics
were also the same as used in the previous studies
except that a log amplifier was added to accommo-
date both the small off-axis scattering and the
large direct transmission signals. The energy de-
tected output was sampled by the computer and re-
corded in digital form along with appropriate posi-
tional information for statistical analysis and
display.
A typical scattering experiment consisted of
sixteen angle scans taken at 2.5 mm increments
along the axis of the cylindrical sample. In each
angle scan, data was collected at 163 equally
spaced increments of 2° from a scattering angle of
-153° to 165°. The transmitted tone burst was
10 MS long while the receiver gate was 6 ys long
and centered in the scattering volume.
Fresh samples of pig liver and human liver
samples obtained at autopsy have also been studied
ul trasonical ly by angle scanning and varying fre-
quency. Cylindrical plugs were taken from peri-
pheral regions near the capsule to minimize the
inclusion of large vessels. Transmit pulse length
and receiver gates were set as in the random medi-
um model studies to minimize inclusion of specular
reflectors from the surface.
B. Optical Data
Optical data was acquired under computer con-
trol by sampling a TV image obtained through a
transmission microscope. Drops of the particle
suspensions were spread over a microscope slide
and typical areas digitized. Five different
images from a smear of each particle suspension
were obtained for analysis. The individual images
consisted of 8-bit samples representing light in-
tensity transmission over a field of 1.7 mm span-
ned by 384 x 384 matrix elements.
The cylindrical liver specimens were fixed in
formalin and standard sectioning and staining
procedures used to evaluate histomorphologic chang-
es in the sections that were studied ultrasonical-
ly. Each stained section was digitized into a
256 x 256 matrix covering a field of 4.5 mm. In
the optical studies of liver tissue, a Van Gieson
connective tissue stain was employed to bring out
structures believed to cause ultrasonic scattering
and the computed average power spectra were used
to characterize the architecture for comparison
with ultrasound angle scan data.
The digital representations of the optical
images obtained through the microscope-TV system
were enhanced to improve the contrast between the
particle boundaries and the background. This was
accomplished by forming histograms of original
image amplitudes and using these histograms to de-
velop mappings that spread the original light in-
tensity transmission values over wider ranges.
The far-field diffraction pattern of each optical
image was obtained by a two-dimensional Fourier
transformation implemented using an integer FFT
algorithm. The log magnitude of the resultant
spectra was calculated to facilitate display of a
wide range of spatial frequency amplitudes. Aver-
ages of the log magnitude spectra from the five
sample functions of the same particle size distri-
bution were computed. The spherical symmetry of
this spectral estimate was used to produce addi-
tional smoothing by averaging spatial frequency
amplitudes determined along radii spaced 1° apart.
The results were used to find average nearest
neighbor distances.
4. Results
Data was collected by frequency scanning 10 mil
monofilament nylon cylinders in arrays with spac-
ings of 0.72 and 1.52 mm at a range of 11.0 cm
and a Fourier transform was applied to infer the
spacings (fig. 1). The influence of transmitter
range on measurements of diffraction by an array
spacing of 1.52 mm was observed for ranges of 4.5
and 11.5 cm (fig. 2). The distances (0.78 and
1.55 mm) indicated by the peaks on the transformed
data in each of the measurements are within 4 per-
cent of the actual values. The envelope of the
data collected at 4.5 cm is different from the far-
field measurements, but did not invalidate calcula-
145
Fig. 1. Swept- Frequency Diffraction by Arrays. Measured values of scat-
tered energy as a function of frequency are shown (left) on a
linear scale with their corresponding Fourier transforms (right)
also on a linear scale for array spacings of 1.52 mm (top) and
0.76 mm (bottom) at a transmitter range of 11.0 cm. The ultra-
sonically determined spacings were 1.55 mm and 0.78 mm respec-
tively.
Fig. 2. Swept-Frequency Diffraction as a Function of Range. Measured
curves of energy as a function of frequency are displayed (left)
on a linear scale with corresponding Fourier transforms (right)
also on a linear scale for transmitter ranges of 4.5 cm (top)
and 11.5 cm (bottom) for an array spacing of 1.52 tom. The ultra-
sonically determined spacings were 1.48 mm at the far range and
1.55 mm at the near range.
146
Fig. 3. Frequency-Dependent Angular Scattering by a Random Medium Model.
Measurements of average intensity are shown (left) on a logarith-
mic scale along with calculated scattering (right) on the same
scale for a frequency of 3.8 MHz (top) and 6.0 MHz (bottom). The
measured data obtained from a suspension of particles demonstrates
more forward scattering as frequency increases and compares well
with computations of scattering from a cloud in which the average
particle radius is 128 pm.
tions of spacings by Fourier inversion of the data.
Angle scans were carried out on three suspensions
of different average size Sephadex2 particles. Mea-
surements of scattered ultrasound signal as a func-
tion of angle were averaged for each distribution
of particle size. The averaged angle scan data is
an estimate of the spectral power in wave number
or reciprocal space of the variations in acoustic
index of refraction.
Averages of ten angle scans for each of three
particle size distributions were obtained. The
transmitted main beam was subtracted from the aver-
age data and the resulting angular scattering data
smoothed to remove artificial fluctuations arising
from averaging only a finite number of sample func-
tions. The final result was compared with calcula-
^Trademark of Pharmacia Fine Chemicals.
tions of scattered intensity using eq. (3) to de-
termine average scattered intensity as a function
of angle at two frequencies for the coarse grade
of particles which had the largest average size
(fig. 3). The averages for two other particle
distributions show how off-axis scattering increas-
es as average particle size decreases (fig. 4).
Enhanced optical transmittance images of five
sample functions and the corresponding average log
magnitude spectra were obtained from smears of
three particle grades (fig. 5). A smoothed spatial
frequency amplitude distribution was found by aver-
aging the amplitudes along radii from which the
nearest neighbor distances may be determined
(table 1).
Pilot studies of acoustic scattering from human
and pig liver tissue by angle scanning have been
carried out (fig. 6). The data shows that there
147
Fig. 4. Size-Dependent Angular Scattering by a Random Medium Model. The
polar plots and corresponding Cartesian displays show average in-
tensity on a linear scale at 5.9 MHz for distributions of scatter-
ers with an average particle radius of 93 ym (top) and with an
average radius of 84 m (bottom). The increase in omnidirectional
scattering for smaller scatterers is evident in both plots and
also demonstrated by the mean scattering angle defined by arrows
crossing standard deviation bars below the Cartesian plots.
Table 1. Nearest neighbor spacing determined
by Fourier transform analysis of
optical transmi ttance.
Particle Mean spacing Spread of
distribution (pm) spacings (+la)
(ym)
G50M 93.3 63.2 - 177.6
G50F 84.3 57.3 - 159.4
is significantly more scattered energy at small
angles than in the backscattered mode. Scattering
In the forward direction becomes greater as the
frequency increases. Optical data obtained from
these specimens shows the well-defined regular
lobular structure of pig liver and the irregular
structure of a post-necrotic cirrhotic human liver
(fig. 7). The power spectra and their correspond-
ing average radial distributions show the same
trend of increased omnidirectional scattering from
pig liver that the acoustic data demonstrate.
5. Discussion
The measured scattering from arrays of nylon
cylinders yields element spacings that correspond
well to actual values and shows the influence of
near-field variations of transducer beams on the
determination of scatterer spacing. The number of
peaks in the transformed data indicates diffraction
by integer multiples of the basic spacing and are a
measure of the number of scatterers in the beam or
the effective width of the common scattering volume
which contains more array elements as the spacing
decreases.
Diffraction-based measurements of array element
spacing as a function of range from the transmitter
show peaks that do not vary in position when the
scatterers are in the near-field of the trans-
ducers. These measurements also support the utili-
ty of a model that employs the far-field assumption
for inferring structure even when the far-field as-
sumption is not strictly satisfied.
Acoustic scattering from Sephadex was measured
using a wavelength that ranged from 2 to 30 times
the particle diameters and corresponds to a transi-
tion region between point scattering and diffrac-
tion by individual spheres. This relationship of
wavelength and particle size was chosen for its poten-
tial similarity to situations of interest in medical
ultrasound. Differences in scattering due to par-
ticle size are still demonstrated by the analysis
of both the acoustic and optical data. The acoustic
data from the Sephadex experiments agrees with the
predictions of the expression describing scattering
from a cloud of particles.
For the coarse grade of particles, the computed
results agree well with the experimental data.
The mean value size parameter of 128 ym is well
148
Fig. 5. Optical Image Analysis of Random Media Model. In (a) particles
with an average size of 84 pm (left) and 93 ym are shown with
the central field enhanced by amplitude mapping. The power
spectra (b) obtained by averaging showed diffraction rings with
an average radial distribution of energy (c) that is inversely
proportional to scatterer size and consistent with the acous-
tic scattering results.
within the range of the manufacturer's specifica-
tions on sizes and is also confirmed by the optical
determinations. Both experimental data and calcula-
tions show that the scattering becomes greater in
the forward direction as frequency increases. Polar
plots of average acoustic intensity as a function
of scattering angle also show that a distribution
of small particles scatters relatively more energy
at larger angles than does a distribution of large
particles.
These results are in agreement with theoretical
predictions about scattering off-axis as a function
of particle size, and support the swept-frequency
studies in which small particles scattered more
energy at high frequency than did large particles
reported earlier [4]. The experiments are compar-
able because the Sephadex suspensions are isotropic
so that angle scanning produces the same result as
frequency scanning.
The acoustic scattering data was collected under
conditions that reasonably approximate the plane
wave, point receiver, and weak scattering assump-
tions employed in the theoretical development. The
170 microseconds between transmit and receive gates
correspond to transducer distances of 12.6 cm from
the center of the scattering volume. This distance
normalized by the ratio, the square of transducer
radius to the wavelength, is 10.9 and is well with-
in the far-field resulting in a plane wave incident
on the scattering volume. The distance also results
149
Fig. 6. Comparison of Average Angular Scattered Intensity From Normal Pig
Liver and Cirrhotic Human Liver. Specimens studied at 3 MHz (upper
panels) show that the normal pig liver scatters more omnidirectionally
than cirrhotic human liver and that, as the frequency is increased to
6 MHz, the forward scattering component increases in both cases. This
is consistent with scattering from collagen which is arranged in a
smaller, more uniform matrix in the pig liver than in the cirrhotic
human liver.
in a maximum phase variation of 16.4° across the
face of the receiving transducer for a wave origi-
nating from a point at the center of the scattering
volume and thus a point receiver approximation is
justified. .
The logarithmic compression of the received
ultrasound signal near 0° scattering angle, needed
to allow simultaneous demonstration of scattering
signals at other angles, indicates that the assump-
tion of weak, single scattering or the Born ap-
proximation is valid.
The polar plots do not demonstrate symmetric
scattering about each side of the 0° scattering
angle beam as anticipated from the isotropy as-
sumption. A reason for this is that a slightly
different scattering volume was sampled on each
side of 0° due to error in concentricity of the
transducer arms.
Optical characterization of a distribution of
spheres shows a ring which agrees with theory.
The good agreement between theory and final ring
structures obtained by processing only five samples
supports the notion that the particle distribution
used for a random medium model is isotropic. Since
the particles are closely packed, the spacings ob-
served by light transmittance analysis are equal
to particle size.
Parameters for the particle distributions de-
rived from averages of radial lines in the two-
dimensional log spectra are in excellent agree-
ment with the values supplied by the manufacturer
as well as being in qualitative agreement with
analysis of acoustic scattering data.
The angle scan data from pig and human liver
specimens demonstrate differences as a result of
the different architecture of the tissues. The
pig liver scatters more omnidirectionally than the
human liver. At lower frequencies, the pig liver
also shows a lobe structure indicative of local
order in the lobular arrangement, while the human
liver scattering, which is predominantly in the
forward direction, results from larger, more ir-
regular spacings. The increase in forward scat-
tering at high frequencies as in the Sephadex ex^
150
Fig. 7. Optical Image Analysis of Liver. In (a), normal pig liver (left) and
cirrhotic human liver have been stained to show connective tissue as
dark bands. The power spectra (b) obtained by averaging show that the
pig liver produces a broader distribution of scattering than the human
liver. The relatively isotropic patterns have been reduced to a one-
dimensional form (c) by averaging radial lines at 1° increments to produce
data comparable to ultrasound angular scattering when viewed as a Carte-
sian plot. The optical analysis demonstrates the same increased spread
in scattering from pig liver that the acoustic patterns contain.
151
periments indicates that important scattering may
take place from elements in the size range includ-
ed in the Sephadex studies. The data also implies
that backscattered energy, while easier to monitor
in practical situations, represents only a small
amount of tissue structure information. It re-
mains to determine whether observed peaks in the
scattering pattern are due to many randomly spaced
scatterers in the tissue or a few large single
scattering objects.
Further investigation is also required to deter-
mine the specific structural changes observed
acoustically, correlate these results with histo-
pathology, and establish the reproducibility of
the results from these initial studies of liver
specimens .
6. Conclusion
Experimental observation of acoustic diffraction
by arrays using swept-frequency ultrasound have
agreed well with theoretical predictions. The ex-
periments indicate that near-field beam nonuniformi-
ties need not invalidate inference of scatterer
spacing from a model employing a far-field approxi-
mation. However, more results are needed to bound
limitations imposed on measurement of scattering
by acoustic diffraction as a result of beam non-
uniformities and misalignments.
The ultrasound angle scan data from Sephadex
shows size-dependent scattering and agrees with
theoretical predictions as well as with optical
analysis. The Fourier analysis of optical data
accurately confirms the particle size distribution
information provided by the vendor and demonstrat-
es a way optical images can be quantitatively
analyzed to provide a reference for comparison with
ultrasound data. The results obtained indicate the
potential for remote probing with ultrasound to
characterize a random medium from its scattering
and the use of digitally processed optical data
to confirm the results.
Preliminary data from swept-frequency and angle
scan experiments with pig and diseased human liver
tissue shows the importance of scattering at small
angles and also that this technique may lead to a
useful diagnostic tool. However, additional data
is needed to characterize tissue and thus point
the way to optimum parameters for clinical deter-
minations of tissue structure from diffraction ef-
fects with ultrasound.
Acknowledgments
We wish to acknowledge important assistance in
the conduct of this work by Edward E. Eyler, C.
Robert Hoffman III, and Jeffrey Astheimer who de-
veloped computer programs, and Frank H. Slaymaker
and Peter H. Helmers who designed major parts of
the mechanical and electronic hardware used in
these studies.
This work was supported by the National Science
Foundation under grant #APR75-14890 and the Nation-
al Heart and Lung Institute under grant #HL15016.
[2] Chivers, R. C, Hill, R. R. , and Nicholas, D,,
Frequency Dependence of Ultrasonic Backscatter-
ing Cross-Section: an Indicator of Tissue
Structure Characteristics, in Proc. of the 2nd
World Congress on Ultrasonics in Medicine,
Rotterdam, The Netherlands, 4-8 June, 1973,
M. deVlieger, D. N. White, and V. R. McCready,
eds., pp. 300-303 (Excerpta Medica, Amsterdam,
1974).
[3] Heyser, R. C. and LeCroissette, D. H., A new
ultrasonic imaging system using time delay
spectrometry. Ultrasound in Medicine and Bi-
ology 1, (2), 119 (March 1974). ~
[4] Waag, R. C. and Lerner, R. M. , Tissue Macro-
Structure Determination with Swept-Frequency
Ultrasound, in Proc. of 1973 Ultrasonics
Symposium, Monterey, Calif., IEEE Cat. No.
73 CHO 807-8 SU 63-66, 5-7 November, 1973.
[5] Namery, J. and Lele, P. P., Ultrasonic Detec-
tion of Myocardial Infarction in Dog, in
Proc. of 1972 Ultrasonics Symposium, Boston,
Mass., IEEE Cat. No. 72 CHO 708-8 SU: 491-
494, 4-7 October, 1972.
[6] Waag, R. C, Lerner, R. M. , and Gramiak, R. ,
Swept-Frequency Ultrasonic Determination of
Tissue Macrostructure, in Ultrasonic Tissue
Characterization, M. Linzer, ed.. National
Bureau of Standards Spec. Publ. 453, pp. 213-
228 (U.S. Government Printing Office, Washing-
ton, D.C., 1976).
[7] Tatars kii, V. I., Wave Propagation in a Tur-
bulent Medium (McGraw-Hill Book Company, New
York, 1961).
[8] Lee, P. P. K., Waag, R. C, and Hunter, L. P.,
Swept-frequency diffraction of ultrasound by
cylinders and arrays, J. Acoust. Soc. Am. 63
(2), 600-606 (1978).
[9] Morse, P. M. and Ingard, K. U., Theoretical
Acoustics, Chapter 8, p. 439 (McGraw-Hill ,
Inc. , New York, 1968).
[10] Goodman, J. W., Introduction to Fourier Op-
tics (McGraw-Hill Book Company, New York,
1968).
[11] Azaroff, L. V. et al., X-Ray Diffraction
(McGraw-Hill Book Company, New York, 1974).
[12] Slaymaker, F. H., Ultrssonic diffraction ap-
paratus users manual, (Preliminary) Progress
Report on Tissue Characterization with Ultra-
sound (NSF Project No. APR 75-14890), Ap-
pendix I, Dept. of Electrical Engineering,
College of Engineering and Applied Science,
Rochester, New York, June 1976. NTIS
#PB267397/AS.
References
[1] Baum, G., Quantized Ultrasonography, in Recent
Advances in Diagnostic Ultrasound, E. Rand,
ed. (Charles C. Thomas, Publisher, Spring-
field, 111., 1971).
152
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed . , National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
QUANTITATIVE MEASUREMENTS OF SCATTERING OF ULTRASOUND BY HEART AND LIVER
J. M. Reid and K. K. Shung
Institute of Applied Physiology and Medicine
and
Providence Medical Center
Seattle, Washington 98122, U.S.A.
Quantitative backscattering measurements have been made using the substitution method
used to measure the scattering properties of blood previously. We have obtained pre-
liminary experimental values Tor the backscattering coefficients of liver and myocardium,
fully corrected for equipment and transducer parameters as well as attenuation of the
scattering tissue. Results show that scattering for both liver and myocardium is an
increasing function of frequency up to 10 MHz. This increase indicates that the tissue
elements responsible for the scattering must be less than 30 micrometers in diameter.
The level of scattering from liver tissue indicates that if it is isotropic scattering
the scattering loss is lange enough to account for about 28 percent of the total at-
tenuation reported at 10 MHz frequency.
Keywords: Absorption; attenuation; heart; liver; scattering; scattering cross-section.
1. Introduction
The recent interest in characterizing biologi-
cal tissues through noninvasive ultrasonic means
has resulted in investigations and reports on
values of velocity, attenuation, impedance, and
scattering and echo strength. The first three
quantities are nearly always measured in a quanti-
tative sense, the last two are usually measured
in a relative sense. The last two quantities are
sometimes spoken of interchangeably as if they were
the same thing. However, echo strength is the re-
sult of the operation of a number of factors:
1) the absolute scattering strength of the tissues;
2) the absorption of the tissues (including the
tissue that is doing the scattering); and 3) many
equipment parameters.
Although it may appear to be expeditious to do
characterization on the basis of echo amplitudes
produced by existing equipment, we believe it
preferable to obtain the data under conditions
which allow the absolute scattering coefficients
to be calculated independent of tissue absorption
and equipment parameters. Knowledge of absolute
scattering parameters is useful for the identifica-
tion of mechanisms and the prediction of possible
applications in medicine. An even more exciting
use, however, is in the design of new equipment
utilizing new approaches to tissue characteriza-
tion which can generally only be done if the data
obtained are independent of the taking equipment.
Quantitative backscattering measurements can be
made using a substitution method which has already
been used to measure the scattering properties of
blood [l-4]i. We have used this method to obtain
preliminary experimental values for the scattering
properties of liver and myocardium. Further more
extensive studies on human tissues will be con-
ducted. Animal tissues were used for coneenience
in adapting the system to solid tissues.
Results have been presented by Nicholas [5] on
backscattering coefficients per solid angle for
liver, spleen and brain. No details are given on
the formalism used to derive the backscattering
coefficient. Nicholas' values for liver are about
an order of magnitude greater than our values, and
at the higher frequencies would predict a scatter-
ing loss which equals the entire observed attenua-
tion. Since this a relatively new field, we be-
lieve it essential to give as many details of the
measurement and data reduction process as possible
and, in this article, will present our complete
procedure.
2. Method
The backscattering coefficient defined as power
scattered per unit solid angle in the backward
direction by a unit volume of scatterers was cal-
culated by comparing the RMS value of the backscat-
tered wave to that of a known reflected wave. The
coefficient can be defined to be independent of
the measuring system and the attenuation charac-
teristics of the tissue [1,2].
Measurements were performed in a large water
^Figures in brackets indicate literature
references at the end of this paper.
153
tank which was filled with normal saline solution.
The temperature of the saline was kept at 18 °C +
1 °C. The samples were fixed on a sample holder
about 15 cm away from the transducer. Excised
calf liver and heart were obtained immediately
after the sacrifice of the animal and stored in
normal saline solution at a temperature of 4 to
6 °C. Measurements were carried out within three
days.
Five liver specimens and five heart specimens
were used with each cut into sample regions which
were examined so that only those samples free of
gross connective tissue or blood vessel interface
were used for the measurements to better define
the bulk properties of each tissue type. The
samples were 2 cm x 2 cm squares with a thickness
of 3 to 5 cm. The liver samples were examined in
a number of randomly chosen orientations so that
30 to 45 experimental values were calculated for
each point on the results. In the measurements of
myocardium samples, the direction of the sound beam
was always perpendicular to the muscle fibers.
Ultrasonic transducers (Panametrics ) with
resonant frequencies at 3, 5 and 10 MHz and a diam-
eter of 0.635 cm were used. Electronic equipment
consisted of a gated burst generator, pulse echo
receiver, and a controllable receiver gating cir-
cuit [2,3]. All time measurements were made on an
oscilloscope and the output gated echo measured by
a true RMS voltmeter. A calibrated attenuator, im-
pedance matched to the receiver input circuit, was
adjusted to obtain the same reference reading on
the voltmeter for the reflected wave and the scat-
tered wave.
The backscattering coefficient was calculated
from the attenuator readings, equipment and equa-
tions expressing the received available power from
targets located on the axis in the far field of a
round piston transducer [1]. The received avail-
able power in the wave reflected from a plane in-
terface can be written
where nd = backscattering coefficient per unit
solid angle, per cm^
S = cross-sectional area of the sound beam,
cm2
o = attenuation coefficient of the scatter-
ing tissue
c = velocity of sound in the scattering
tissue
R = range to surface of scattering tissue
ti = time from first surface echo to open-
ing of receiver gate
ta = time from first surface echo to closing
of receiver gate
Equation (2) neglects the front surface reflection
of the sample since the cut specimens we employed
did not include surface covering membranes, the
usual source of such reflections. The terms in
brackets are correction factors to account for
the total attenuation losses of the scattering
material. These losses include absorption of the
sound in the scattering tissue between its sur-
face and the region from which scattered echoes
are gated into our receiver and measuring circuit,
absorption of sound during the transmitted pulse
and interaction factors which depend on the absorp-
tion of the wave during the duration of the re-
ceiver gate and the transmitted pulse. The scat-
tering coefficient can be found by dividing eq.
(2) by eq. (1) and solving for the coefficient.
The resulting equation is:
E^A^xe
.4amR
0
R2Tp4x2
k2
(1)
where = received available electrical power (RMS)
Pq = transmitter electrical power (RMS)
E = transducer efficiency (a function of
frequency)
A = transducer area
R = range from transducer to interface
Tp = pulse repetition period
T = transmitted pulse length
X = wavelength
am = pressure attenuation coefficient for
propagation medium
k = amplitude reflection coefficient between
propagating medium and reflector
In a similar manner, the available power in a wave
scattered from a distribution of lossy scatterers,
Pg, can be written [2]:
E2A2S2e
.4amR
P 8R'*TDX2a2c
0 P
(2)
nd(e
-2act
aCT _aCTK
- e )
2k2R2a2cT
(3)
P s(e-2°'cti _ g-2act2)(gaCT _ g-acx.
The ratio of P^ to Pp is obtained from the
change of the input attenuator settings necessary
to make the RMS readings of the pulse from the
flat reflector equal the reading of the gated,
scattered waves. Theoretically, this ratio should
be expressed in terms of the open circuit trans-
ducer voltages due to the scattered and the re-
flected waves. When the attenuator is connected
we read instead a loaded or closed circuit voltage.
Since we use the same transducer and receiver im-
pedance for the two waves, the loaded voltage is
less than the open circuit voltage by a complex
frequency-dependent factor which depends on the im-
pedances in the circuit. This factor is the same
for the scattered wave as for the reflected wave,
however, and hence divides out when the ratio is
taken. Therefore, the ratio is the same for the
open-circuit as for the loaded condition. The
other factors in eq. (3) can be obtained by direct
measurement of apparatus dimensions, oscilloscope
time settings and separate measurements of attenua-
tion and velocity within the scattering tissue.
In this investigation, these tissue properties
were taken from the published literature [6-8].
The time lapse between the receiver gate and the
tissue - saline surface (tj is in the order of 10
to 20 \im. Attenuation losses in this layer as well
as losses in the sample volume are corrected for
by use of eq. (3). The amplitude reflection coef-
ficient between the saline and tissue surface was
less than 0.1 and assumed negligible.
Figure 1 is the data collection sheet filled
out by the experimenter at the time of measurement
for one specimen at one frequency. The section of
the sheet before the calibration run consists of
equipment parameters. The calibration run allows
154
Date * MARCH, 1977
Data Collection Sheet for the Scattering Experiment
Sign_
Specimen: CALF MYOCARDIUM
Wave Frequency: 10 MH2
Temperature of Water Bath: 18 "C
Gate Width: 10 us T^: 1(
Transducer Position: 34 cm
Sample Position: 52.7 cm
Beam Width: 0.27 cm
Calibration Run:
Reflector Position
Pulse Repetition Rate; 1 kHz
Pulse Width: 4 vs
Reflector Positi
Beam Cross -Sect ion : 0.057 cm^
Attenuator Readi
58
Reference on RMS Meter: 0.003 V, -8 dB
RCDistance between Transducer and Sample) = 15.4
a [Attenuation Coefficient for Sample) = .1
c(Velocity of Propagation) = 1.58
DATA
1 11.5 2 10 3 11 4 9,5
89 9 8 10 10 11 9.5
IS 11
REDUCED DATA
1 0.017 2 0.023 3 0.015 4 o.Oll
8 0.01 9 0.008 10 0.012 H 0.011 1
15 0.015
Average :
5 10.5
12 15
cm -31.19
0.29
Nepers/cm 1,68
-20.59
cm/second -29.19
14 15
SCATTERING COEFFICIENT
0.014 6 0.03 7 0.06
0.038 13 0.06 14 0.038
Fig. 1. Sample data collection sheet. For each
specimen at each frequency the data shown
were recorded by the experimenter.
us to check for R"2 echo dependence, eq. (1), as
proof of far-field conditions, and relates the at-
tenuator reading at the surface of the sample to
the attenuator reading at a reference position with-
in the tank. Readings from the reference position
are used to monitor system gain. The next four
lines retain data which apply to this particular
experiment. The distance, R, shown was calculated
from the position of transducer and sample carriers
shown at the top of the page and the use of correc-
tions for the distance between the active surfaces
of transducer and sample and the holder position.
Attenuation coefficients and velocity of propaga-
tion (inadvertently recorded as cm/second rather
than meters/second used in calculation) were ob-
tained from the literature. The numbers shown at
the right side of the figure were notes made during
calculation in which various factors in eq. (3)
were expressed in terms of dB. Under data we list
the attenuator readings necessary to keep the RMS
meter at its reference reading.
The result of final calculation of scattering
coefficient, nd, is shown at the bottom of the
page.
3. Results and Discussion
The experimentally determined values of scatter-
ing coefficient are plotted in figures 2 and 3.
0.03
0.02
0.01
0.005 -
0.001
^ 0.0005
0.0001
For Myocardium
O Normal
• Fixed
till
1
2 3 4 5 6
Frequency (MHz]
10
Fig. 2. Measured backscattering coefficient for
the myocardium as a function of frequency.
Experimental points are the average of
30 to 45 separate measurements on five or
six samples. Bars indicate the standard
deviation vs_. frequency slopes for Rayleigh,
fourth power scattering and for third power
scattering.
One point using tissue fixed with 10 percent for-
malin is included for comparison. The data show
that liver is a much stronger scatterer than heart
tissue at the lower MHz frequencies. Also, the
fixed tissue has a stronger scattering than the
normal tissue in each case. The backscattering
from both myocardium and liver increases with fre-
quency. Figure 2 shows reference lines drawn
through arbitrary points to give the attenuation
slope for both fourth power (Rayleigh) and third
power scattering. At the higher frequencies, the
heart tissue appears to be approaching approximate
fourth power dependence and the liver tissue less
than a third power dependence. It must be recog-
nized that the program is still in its early stages
since only a few data points have been taken and
positive conclusions, particularly with respect to
frequency dependence, are not warranted. Scatter-
ing does appear to increase with frequency, how-
ever.
Since the scattering at 10 MHz is still increas-
ing with frequency (fig. 2), it appears that the
155
0.03
0.02
0.01
01
S 0.005
0.001
For Liver
A Normal
▲ Fixed
1
1
2 3 4 5
Frequency (MHz)
Fig. 3,
Backscattering coefficient for liver tis-
sue as a function of frequency.
structure responsible for producing the scattering
must be smaller than the wavelength. For a spheri-
cal particle it is known that the circumference
must be less than one wavelength, indicating that
the diameter of such tissue elements in liver is
less than 30 microns.
From these data, it is possible to estimate the
contribution to attenuation caused by scattering
alone. In cases where the angular dependence is
known the total scattering loss, as, can be found
from
4tt
|i ,o)da)
fluence the total attenuation. Since the scatter-
ing contribution is a strong function of frequency,
we might expect that at a sufficiently high fre-
quency the attenuation of some tissues or tumors
might increase at a rate greater than the first
power dependence found by previous investigators
working at lower frequencies. If verified, such
behavior could have great importance for tissue
characterization, whether it be done using scat-
tered waves or measurements of attenuation.
References
[1] Reid, J. M. , The Scattering of Ultrasound
by Tissues, in Ultrasonic Tissue Charac-
terization, M. Linzer, ed.. National Bureau
of Standards Spec. Publ. 453, pp. 29-47
(U.S. Government Printing Office, Washing-
ton, D.C., 1976).
[2] Sigelmann, R. A. and Reid, J. M., The analy-
sis and measurement of ultrasound backscat-
tering from an ensemble of scatterers excit-
ed by sine wave bursts, J. Acoust. Soc. Am.
35^, 1351 (1973).
[3] Shung, K. K., Sigelmann, R. A., and Reid,
J. M. , The scattering of ultrasound by blood,
IEEE Trans, on Biomedical Engineering, BME-23,
6, 460 (1976).
[4] Shung, K. K., Sigelmann, R. A., and Reid,
J. M. , The Scattering of Ultrasound by Red
8 10 Blood Cells, In Ultrasonic Tissue Characteriza-
tion, M. Linzer, ed.. National Bureau of
Standards Spec. Publ. 453, pp. 207-212 (U.S.
Government Printing Office, Washington, D.C.,
1976).
[5] Nicholas, D., The Application of Acoustic
Scattering Parameters to the Characterization
of Human Soft Tissues, in 1976, IEEE Ultra-
sonic Symposium Proc, pp. 64-69 (1976), No.
76CH 1120-5SU.
[6] Wells, P. N. T., Physical Principles of Ultra-
sonic Diagnosis (Academic Press, London and
New York, 1969]".
[7] Pauly, H. and Schwan, H. P., Mechanism of ab-
sorption of ultrasound In liver, J. Acoust.
Soc. Am. 50, 192 (1971).
I I I I
(4)
where
Dd
I
0
the solid angle
differential scattering coefficient
the Incident direction of the wave
the direction of observation
As a first approximation, let us assume that the
scattering from liver is isotropic, then.
4Trn
180
At 10 MHz frequency the attenuation coefficient for
liver due to scattering alone is then 0.3 neper/cm.
This loss is about 28 percent of the total attenua-
tion found in the literature [7j~.
This investigation has obtained experimental
evidence that the scattering from within functional-
ly homogenous tissue may be strong enough to in-
[8] Yuhas, D. E., Mimbs, J. W. , Miller, J. G.,
Weiss, A. N., and Sobel , B. E., Changes in
Ultrasonic Attenuation Indicative of Regional
Myocardial Infarction, in Ultrasonics in
Medicine, D. White, ed., Vol. 3 (Plenum Press,
New York, 1976).
[9] Gramiak, R. , Hunter, L. P., Lee, P. P. K. ,
Lerner, P. H., Schenk, E. , and Waag, K. C,
Diffraction Characterization of Tissue Using
Ultrasound, in 1976, IEEE Ultrasonics Sympos-
ium Proc. , pp. 60-63 (1976), No. 76CH 1120-5SU.
^It has to be noted this figure obtained was based
on a very rough assumption. In reality, the
scattering from liver is not isotropic as indicated
by the experimental results of Gramiak et_ aj_. [9].
Our measurements do average n over many directions,
so we feel the conclusions are warranted.
156
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed. , National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
DEPENDENCE OF ULTRASOUND BACKSCATTER FROM HUMAN LIVER TISSUE
ON FREQUENCY AND PROTEIN/LIPID COMPOSITION
M. Freese
Radionics Ltd.
Montreal, Quebec, Canada
E. A. Lyons
Winnipeg Health Sciences Center
Winnipeg, Manitoba, Canada
The dependence of ultrasonic volume backscatter on frequency, lipid and protein con-
tent in normal and fatty human (post-mortem) liver tissue was investigated with a view
towards the possible development of a noninvasive quantitative diagnostic test for fat-
ty liver. Mean values of the backscatter coefficient, its range and frequency de-
pendence were determined for normal liver in the 1 to 5 MHz range and for fatty liver
at frequencies of 2.25 and 3.56 MHz. The results indicate that frequencies below about
2 MHz should be avoided for quantitative measurements of the backscatter in liver.
Although the measurements revealed considerable variation in the backscatter levels of
normal liver, the backscatter levels in moderately and severely fatty liver were sig-
nificantly greater than the normal range. Simple linear correlation of the backscatter
magnitude with the lipid content for 21 samples (10 normal, 8 fatty, 1 cirrhotic, 2
other abnormal) yielded a value of 0.94, significant at the 1 percent level.
Key words: Backscatter frequency dependence; cirrhosis; composition-dependent
scattering; fatty liver; stochastic scattering; tissue characterization;
ultrasound attenuation; ultrasound diagnosis; ultrasonic tissue
scattering.
1. Introduction
Quantitative data on the dependence of ultra-
sound scattering on the physical structure and
composition of tissue is needed to understand the
underlying scattering processes, to aid the de-
velopment of optimum diagnostic scattering
methods and to serve as a reference in clinical
applications. From the standpoint of these ob-
jectives, specifically, the possible development
of a noninvasive diagnostic test for fatty liver
[1]^, this paper describes the results of a
series of quantitative measurements of ultrasonic
volume backscatter in the 1 to 5 MHz range for
normal and fatty post-mortem human liver as a
function of frequency and protein-1 ipid content.
The results suggest that a noninvasive quantita-
tive test for fatty liver based on the magnitude
of the backscatter is feasible if the in vivo mea-
surement problems can be overcome.
2. Methodology
The pulse echo technique and the calibration
procedure employed for the backscatter measurements
^Figures in brackets indicate literature
references at the end of this paper.
have been described in detail elsewhere [2]. Es-
sentially, the received backscatter from the tissue
is range gated, passed through either a square-law
or linear envelope detector and integrated; time
varied gain is employed to compensate for sample
attenuation. The measurements are referenced to
stainless steel ball bearing targets of known cross-
section.
A. Backscatter Parameters
The parameters used to describe the backscatter-
ing are the average backscatter coefficient <n>
and the average "envelope" backscatter coefficient
<r>, the brackets denoting ensemble averaging. For
monochromatic (CW) frequencies, <fi> is identical to
the average backscatter cross-section per unit
volume, while for pulses, the difference between
<fi> and the corresponding CW value is dependent on
the absorption coef f icient--path length--pulse
bandwidth product and the scatterer frequency
responses [2,3]. (The maximum estimated error due
to the finite pulse bandwidth was approximately
0.1 dB in these measurements.) The envelope back-
scatter coefficient <r> is proportional to the
average magnitude of the backscatter signal but is
normalized to be consistent with the statistical
relation <fi> = <v'^>.
157
B. Signal Processing
Stochastic signal process theoretical results
were employed to determine appropriate thresholds
for refining of the raw data. Scattering by the
larger blood vessels in the liver poses a problem
because its contribution tends to be largely extra-
neous to the scattering process of interest. To
reduce this interference, the raw data were refined
by rejecting individual measurements with values of
the coefficient of variation y > 0.81, the coeffi-
cientjOf variation being defined as y = {<r'^> -
<r>2)^/<r>. The presence of isolated large specu-
lar echoes tends to increase the expectation value
of Y (= 0.52 for a random process described by a
Rayleigh distribution [4]). Ordinarily, the proba-
bility of Y exceeding 0.81 is about 7 percent. The
A-scan time averaging employed reduced this proba-
bility to 5 percent at 1.1 MHz decreasing to 0.5
percent at 3.56 MHz.
To further improve the averaging and avoid the
long delay path required for far-field measure-
ments, the measurements were performed in the near
fields of the transducers. The measured values
were then converted to equivalent far-field values
using the experimental results obtained by Freese
[5]; the same transducers were used in the present
measurements^. The near- to far-field conversion
factors ranged from + 1.4 and + 1.1 dB at 1.1 MHz
to - 1.1 dB at 4.88 MHz.
Time varied gain based on running mean absorp-
tion estimates were employed during the measure-
ments with appropriate corrections based on the mea-
sured absorptions being applied at the completion
of the experiments [6].
C. Measurements
The ultrasonic measurements on the liver samples
were conducted with two exceptions within one to
three days post-mortem. The samples were removed
from the cadaver prior to any postmortem infusion
procedures, and were cut from the anterior portion
of the liver extending below the ribs, with the
backplane of the sample sliced roughly parallel to
the external liver surface. With the samples im-
mersed in physiological saline held at 20 °C and
the ultrasonic beam incident perpendicularly on the
external liver surface, the backscatter was measur-
ed at two depths extending from 0.6 cm to 1.8 cm,
and from 1.2 cm to 2.4 cm below the surface, re-
spectively. Four separate measurements were made
at each depth. The total insonified volume averag-
ed about 10 cm^ per sample (5 cm^ per sample in the
case of the 1.1 MHz measurements).
The absorptions were measured by means of a sub-
stitution technique. To the extent possible, all
of the measurements were carried out under double-
blind conditions.
D. Diagnostics and Biochemical Assays
Subsequent to the ultrasonic measurements, the
samples (nineteen male, six female) were classified
2The conversion factors (as a function of trans-
ducer distance) were determined from scattering
measurements on model media containing random
scatterers whose dimensions were comparable to
tissue cells. Corresponding data obtained for
whitefish muscle tissue were generally found to
be within 0.5 dB of the model media values.
on the basis of clinical and autopsy data as either
normal or abnormal. The abnormal group (eight male,
four female) was then further categorized accord-
ing to the lipid content--fatty (seven male, two
female) or nonfatty. One normal and one fatty
sample were rejected because of technical problems.
The lipid, protein, and moisture contents of the
samples were determined by B. Guy Hunt Laboratories
(Winnipeg) using standard chemical analytical
methods .
3. Results and Discussion
The average lipid content and physical data for
the normal samples (table 1) are in good agreement
with generally accepted values [7]; however, the
protein content appears to be about 20 percent low.
Detailed data on the abnormal samples is given in
table 2.
Table 1. Composition and physical characteristics
of normal liver samples. ^
Factor Mean Standard
deviation
Total lipid {%)^ 4.8 1.1
Protein {% nitrogen x 6.25) 14.6 2.1
Moisture {%) 77.5 2.8
Liver weight (g) 1680 280
Age of deceased (y) 55 26
Weight of deceased (kg) 67 9
^Eleven male, one female, ranging in age from
,15 to 93 years.
Estimated accurace of liqid analytical deter-
mination ± (0.1 {% lipid) + ]%).
A. Normal and Fatty Liver Tissue
Ultrasound Attenuation
The attenuation was measured in eight of the
normal samples, and yielded a mean value of 0.64
dB/cm/MHz for the frequency normalized absorption
coefficient, a/f. Possibly reflecting a lower
protein content, this value is about 10 percent
lower than the mean value quoted by Wells [8] for
the 3 to 5 MHz range. The average attenuation for
the normal samples at the 2.25, 3.56 and 4.88 MHz
frequencies were 1.5, 2.3 and 3.2 dB/cm, respectively.
In the case of the fatty livers, the absorptions
were determined for all but the 7.7 percent lipid
sample (table 2). Excluding the 45.6 percent lipid
sample, which exhibited anomalously high absorption,
the average value of a/f for the fatty livers was
0.86 dB/cm/MHz or about one third greater than
for the normal livers.
The two extremely fatty samples (table 2) were
rather interesting for, although similar in ap-
pearance, they had markedly different absorptions.
The 35.5 percent lipid sample measured 2.74 dB/cm
at 3.56 MHz (six replicates) which is only slight-
ly more than in normal liver, while the 45.6 per-
cent lipid sample measured a remarkable 14 dB/cm
(three replicates). The reason for this anomalous-
ly high absorption is not certain but it seems to
point to the possible presence of gas bubbles in
the latter sample, despite the fact that the liver
sample was obtained one day post-mortem. Also, un-
158
Table 2. Composition and clinical data for the abnormal samples.
Group Total Protein Moisture Clinical information Autopsy results
lipid sex age height weight history Liver Liver
% % % y m kg weight, g appearance
Fatty
7.7
9.0
11.4
14.7
14.9
19.1
35.5
45.6
16.0
14.0
11.4a
13.2
13.2
11.9
10.6
8.4
72.2
71.1
74.3
64.1
71.2
70.1
52.5
44.4
M —
M 26
M —
M 45
M 54
F 69
M 60
M 50
1.63
1.70
1.65
1.80
1.63
1.82
1.68
52
69
64
59
66
84
117
alcohol ism
(blood alcohol
150 mg)
alcohol ism
(blood alcohol
16 mg)
alcohol ism
1630
1690
1940
1900
pale yellowish
alcoholic hepa-
titis, mod. fat
infilt.
1600 pale
3450
5600
yell ow
yellow, fatty
Nonfatty
3.6 18.1
3.7 15.2
3.8 16.8
79.8
79.6
78
M 64
91
76
1.74
1.56
1.51
82 cirrhosis
1730
40 diabetes (diet) 1220
37 — 730
yellow, firm
nodul ar
adv. cirrhosis
smal 1
Estimate.
like the 35.5 percent lipid sample, the 45.6 per-
cent sample was buoyant in the 0.1 N saline solu-
tion.
B. Normal Liver Backscatter Coefficient,
Range and Frequency Dependence
The backscatter coefficients and corresponding
envelope backscatter coefficients for the normal
samples are plotted as a function of frequency in
figure 1(a) and (b). (A number of single data
points are also included.) The ranges of the back-
scatter coefficients at each of the measurement
frequencies are extremely broad. For example, at
2.25 mHz, <^> varies from 0.17 to 2.9 mm2/cm3— a
range of over 12 dB. With the exception of a
single low value at 4.88 MHz, which may have been
the result of a faulty measurement, the largest
range is exhibited by the 1.1 MHz values. It is
tempting to attribute this primarily to first-
order diffraction effects [9,10] and to the great-
er statistical fluctuation at this frequency, but
a more careful analysis of the measurements does
not seem to support this view. While large fluc-
tuations in the individual measurement values were
observed in many of the samples, when the measure-
ments were repeated on the same sample using slight-
ly different aspect angles and insonified sites,
the mean values seldom differed by more than 1 or
2 dB. Furthermore, if we take into account the
dimensions of the blood vessels in terms of wave-
length, the decrease of the highest <fi> values be-
tween 1.09 and 2.25 MHz is consistent with oblique
incidence scattering by cylinders of corresponding
radii in the transition region from Rayleigh to
geometrical scattering (i.e., the vessels will
tend to scatter specularly with increasing fre-
quency). Since the larger blood vessels tend to
radiate towards the liver surface, the oblique
aspect angles would be more probable in our measure-
ments resulting in an effective decrease in the
backscatter at the higher frequencies. In contrast
to the vessel orientations, the orientations of the
liver lobule "fascia" appear essentially random
over the extent of the insonified volumes. This
would lead one to expect the ensemble averaged
backscatter to be independent of aspect angle. In
attempting to answer the question concerning the
relative contributions of the liver lobules, blood
vessels and other liver structures to the overall
scattering process [11], our results would seem to
indicate that scattering by the larger vessel in-
homogeneities, if present in the insonified volume,
will tend to dominate the backscatter below about
1.5 MHz.
The average frequency dependence of <fi> over the
2.25 to 3.56 MHz range is given by f0-8(fig. 1(a)),
decreasing to about f° in the 1.09 to 2.25 MHz in-
terval, and increasing to f^-^ between 3.56 and
4.88 MHz (if we exclude the 4.88 MHz value for the
borderline sample). While there are too few values
to provide more than a rough estimate of the fre-
quency dependence in the latter two intervals,
these values are in good agreement with those ob-
tained by Nicholas [12] using spectral analysis.
The relatively low frequency dependence observed
illustrates the effect that the presence of even a
relatively small number of "geometric! region"
scatterers can have on the composite frequency de-
pendence. For this reason the frequency dependence
of diffuse volume backscatter may not be as aseful
a parameter for some tissue diagnostic applications
as, the magnitude, the angular dependence or the
absorption.
159
10
2J3 3D 5D
Frequency (MHz)
(a)
70
2.0 3JD
Frequency (MHz)
(b)
50
7.0
Fig. 1. (a) Backscatter coefficients as a function of frequency for samples of normal
liver tissue. The logarithmic mean values of <n> for the 2.25 and 3.56 MHz
data are indicated by connecting heavy barred lines. The dashed lines corre-
spond to ± 6 dB about the logarithmic means and bound essentially all of the
measured values in this frequency range. The sample with the lowest back-
scatter coefficients suggests a borderline sample. Its moisture content of
81 percent was second highest of the normal samples,
(b) Corresponding envel ope backscatter coefficients.
Similar comments apply to the envelope backscat-
ter coefficient shovm in figure 1(b) although the
frequency dependence is only half as great (or less
if the scattering process is inhomogeneous) . The
principal advantages of using <r> are that it sim-
plifies the electronics requirements and is less
sensitive to isolated large extraneous echoes
thereby reducing the consequent error. For these
reasons, we will consider mostly <r> instead of
<n> in the following paragraphs. However, to ob-
tain an estimate of <n> one need only compute
(1 + Y^)<r>2 using the average measured value of
0.6 for Y instead of the theoretical value of 0.52.
C. Backscatter Dependence on Lipid and
Protein Content
Values of <r> were compared with the lipid and
protein contents of both the normal and abnormal
samples. Figure 2 shows the 2.25 MHz values of
<r> as a function of the protein content (P) for
normal samples. The value of the simple linear
correlation coefficient rj-p = 0.776 and is signifi-
cant at the 1 percent level (t-test assuming rpp =
0; ten samples). Comparable results were obtained
at the higher frequencies. This implies that in
normal liver inherent differences in the protein
content may result in <r> varying as much as 9 dB.
(Conversely, taking the protein into account and
assuming the scattering processes add in quadrature,
the 12 dB range of the 2.25 MHz values in figure
1(a) and (b) can, in principle, be reduced by
4.5 dB.)
2.0
0.0
Fig. 2
fr = 2.25 MHz
14
Protein
16
(7o nitrogen
20
6.25)
Dependence of <r> on protein content in
normal liver; the correlation coefficient
r^p = 0.776.
160
No significant correlation was observed at 1.1
MHz. This fact coupled with the previous observa-
tions would seem to indicate that frequencies below
about 2 MHz should be avoided for quantitative mea-
surements of volume backscatter from liver tissue.
In contrast to the correlation of the backscat-
ter with protein content, the correlation of <r>
with the lipid content of the normal samples was
not significant. The presence of glycogen and
other minor hepatic constituents (moisture tends
to correlate inversely with lipid and protein) has
been neglected but their contribution is not likely
to be significant in normal liver.
The effect of the lipid becomes apparent in the
fatty liver samples, which ranged from 7.7 to 45.6
percent lipid. The backscatter and lipid content
(L) are highly correlated (rp^ = 0.980 and 0.978
at 2.25 and 3.56 MHz, respectively) and suggest a
roughly linear relationship between the two as
shown in figure 3 for the 2.25 MHz values.
However, in view of the limited number of mea-
surements and the possibility that the 45.6 percent
lipid sample may have contained gas bubbles, the
latter is probably at least partly due to coinci-
dence since a linear dependence for a lipid content
aoproaching 50 percent would seem unlikely.
The histological appearance of mildly and severe-
ly fatty livers are shown in figure 4(a) and (b).
It can be seen from these pictures that the struc-
tural changes that occur in hepatic tissue with in-
creasing lipid content span the spectrum from nearly
normal to grossly abnormal. If we combine the two
Fig. 4. (a) Mildly fatty liver in an alcoholic. Many of the hepatocytes contain large lipid globules
(clear areas in the stained cytoplasm). Magnification 25X.
(b) Severe fatty liver in an alcoholic. The nuclei of the cells are displaced by large lipid
globules; little cytoplasm remains. Magnification 25X.
5,0
4.0
3.0
20
1,0
0.0
20 30
Total Lipid ( % )
40
50
Fig. 3
Dependence of <r> on lipid content in fatty
liver; the correlation coefficient rpL =
0.980; ff = 2.25 MHz.
groups of samples and take the protein content into
account, the multiple linear regression equation
for <r> at 2.25 MHz becomes <r> = -1.4 + O.IOL +
0.12P with rp Lp = 0.966. This indicates that the
effective scattering contributions by the protein
and lipid (either directly or indirectly) are rough-
ly comparable on a percentage weight basis, with
the protein contributing about 40 percent more to
<fi> at this frequency than the lipid for equal per-
161
cent.ages of the two constituents. The correspond-
ing result at 3.56 MHz was <r> = -1.9 + O.IOL +
0.17P with rp |_p = 0.939. In this case, data was
available for'only eight normal and six fatty
samples. Additional measurements will be needed to
establish if the difference in the relative contri-
butions implied by these two regressions (and there-
fore the frequency dependence of the normal and fat-
ty samples) is signif icaitit.
For predictive purposes we require the percent
lipid as a function of <r> and the percent protein.
Moreover, the remaining abnormal nonfatty samples,
listed in table 2, should also be included in the
regression. The resultant linear regression equa-
tion obtained at 2.25 MHz was L = 15.1 + 9.2<r>
-1.2P, with r|_ j,p = 0,971. This regression with
18 degrees of freedom (21 samples) and F-test value
= 157 is significant at the 1 percent level. Com-
paring it to the previous 2.25 MHz regression, the
effect of the additional samples is seen to be
minimal .
5 0 (-
0 10 20 30 40 50
Total Lipid ( % )
Fig. 5 Dependence of <r> on the lipid content
for the combined group of normal and ab-
normal samples; the correlation coeffi-
cient rj,^ = 0.943.
In potential diagnostic applications the percent
protein would most likely not be available. The re-
sultant simple two variable linear regression equa-
tion obtained, L = -442 + 11.0<r>, is graphed in
figure 5. The correlation coefficient = 0.943
is essentially the same as when the nonfatty ab-
normal samples were excluded. Figure 5 shows that
the backscatter coefficients are substantially
greater for the moderately and extremely fatty
livers. The corresponding regression at 3.56 MHz,
based on 17 samples, resulted in L = -6.2 + 11.3
<r> and rL^ = 0.914. In view of the limited number
of samples and the large weighting by the two ex-
tremely fatty livers, these regressions can only be
considered preliminary. Nevertheless, they seem to
suggest that a simple quantitative diagnostic test
based on the magnitude of the backscatter is feasi-
ble in principle.
4. Summary
Experimental values of the volume backscatter
coefficients and estimates of their distribution
and frequency dependence were obtained for normal
liver. These were discussed and compared from the
point of view of the underlying scattering proces-
ses, the liver composition and potential diagnostic
requirements. The results indicate that in the
case of liver, frequencies under 2 MHz should be
avoided for quantitative diagnostic backscatter
measurements.
Significant correlations of the backscatter with
protein content were observed in normal liver and
with both protein and lipid content in abnormal
fatty liver. In the case of the latter, regressions
significant at the 1 percent level were obtained.
Acknowledgment
This work was supported in part by the Depart-
ment of Environment, Freshwater Institute, Winnipeg,
Canada .
References
[1] Wells, P. N. T., McCarthy, C. F., Ross, F. M.
G., and Read, A. E. A., Comparison of A-scan
and compound B-scan ultrasonography in the
diagnosis of liver disease, Br. J. Radiol. 42,
818-823 (1969).
[2] Freese, M. and Hamid, M. A. K. , Lipid Content
Determination in Whole Fish Using Ultrasonic
Pulse Backscatter, 1974 Ultrasonics Symp.
Proc, IEEE Cat. #74 CHO-896 15U, 69-76 (1974).
[3] Freese, M. , Quantitative diagnostic measure-
ments of ultrasound tissue scattering (in
preparation).
[4] Ol'shevskii, V. V., Characteristics of Sea
Reverberation (Consultants Bureau, New York,
1967).
[5] Freese, M. , Comparison of Near- and Far-
Field Measurements of Diffuse Ultrasonic Tis-
sue Volume Backscatter, 1975 Ultrasonics
Symp. Proc. IEEE Cat. #75 CHO-994 45U, 33-
36 (1975).
[6] Freese, M. and Lyons, E. A., Ultrasonic back-
scatter from human liver tissue; its de-
pendence on frequency and protein/1 ipid com-
position, J. Clin. Ultrasound (in press).
[7] Schiff, L., ed.. Diseases of the Liver, 3rd
ed. (Lippencott Co. , New York, 1969).
[8] Wells, P. N. T., Physical Principles of
Ultrasonic Diagnosis (Academic Press, London
and New York, 1969).
[9] Nicholas, D. and Hill, C. R., Acoustic Bragg
diffraction from human tissues. Nature 257,
305 (1975).
162
[10] Lele, P. P., Mansfield, A. B., Murphy, A. I.,
Namery, J., and Senapati, N., Tissue Charac-
terization by Ultrasonic Frequency-Dependent
Attenuation and Scattering, in Ultrasonic
Tissue Characterization, M. Linzer, ed..
National Bureau of Standards, Spec. Publ.
453, pp. 167-196 (U.S. Government Printing
Office, Washington, D.C., 1976).
[11] Gramiak, R., Hunter, L. P., Lee, P. P. K.,
Lerner, R. M. , Schenk, E., and Waag, R. C,
Diffraction Characterization of Tissue Using
Ultrasound, 1976 Ultrasonics Symp. Proc,
IEEE Cat. #76 CH1120-5SU, 60-63 (1976).
[12] Nicholas, D., The Application of Acoustic
Scattering Parameters to the Characteriza-
tion of Human Soft Tissues, 1976 Ultrasonics
Symp. Proc, IEEE Cat. #76 CH1120-5SU, 64-
69 (1976).
163
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed.. National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
ULTRASOUND BACKSCATTERING FROM BLOOD: HEMATOCRIT AND
ERYTHROCYTE AGGREGATION DEPENDENCE
M. Hanss and M. Boynard
Laboratoire de Biophysique
UER Exp^rimentale de M^decine et de Biologie
74, rue Marcel Cachin - 93000 Bobigny
and
UER Biom^dicale des Saints-Peres
45, rue des Saints-Peres - 75006 Paris
The ultrasound back-scattering of blood has been shown to be increased when
the erythrocytes sedimentation rate is high. Moreover, in this case, temporal
fluctuations of the back-scattered intensity have been also demonstrated. To
explain these results, a simplified theory of ultrasound back-scattering by
blood is presented. The model of the blood structure used is: the erythrocytes
are associated to form spherical transient aggregates having the same cell
number, m. The acoustic wave is scattered by these clusters called "scattering
units." The scattered intensity is calculated for one scattering unit taking a
spherical elastic model, then for a blood unit volume, when the hemotocrit H is
very low. It is found that the back-scattered intensity is proportional to H
and to m. When H is not negligible, a simplified statistical theory is proposed
The main result is that the back-scattered intensity is still proportional to m
but that it passes through a maximum value when H get near 0.3. An explicit
relation between the mean relative value of the intensity fluctuation, H and m
is given. Therefore, the aggregation state of blood can in principle be
determined through ultrasound scattering studies. Further theoretical develop-
ments are in progress so as to take into account the distribution of the number
of erythrocytes in each kinetic units and the variation of this number with H.
Key words: Blood erythrocytes aggregation; blood hemotocrit; ultrasound back-
scattering from blood.
When an ultrasound echographic probe is placed
on top of a vertical cylindrical tube filled with
blood, it has been found that the echo pattern of
the sedimenting blood varies according to its
sedimentation rate [1]^. For high sedimentation
rate samples, the mean back-scattered intensity
is larger; furthermore the echo amplitude at a
given depth in the erythrocyte column presents
characteristic temporal fluctuations. We wish to
present a simple theoretical approach in order to
explain these phenomena. Though the scattering
of ultrasound by blood is increasingly used in
clinical medicine, the fundamental data on this
property are very scarce, and mainly due to Reid
and co-workers [2-5]. Fluctuations (temporal and
spatial) in the echo amplitude have also been re-
ported by Atkinson and Berry [6]. However, none
of these studies can give a straightforward expla-
nation of our experimental results.
^Figures in brackets indicate literature
references at the end of this paper.
Blood Model
The red blood cell (RBC) concentration is
usually defined by the hematocrit H, that is, the
time average of the RBC volume fraction in a given
blood volume. For a normal blood sample, H is
rather large (H = 0.45) so that the RBC interact
each with the other. We will describe these in-
teractions by assuming that m RBC are mechanical-
ly correlated during a period tc larger than the
ultrasound pulse, giving rise to a transient ag-
gregate which we assume to be spherical, with
radius A; this dimension could be thought of as a
correlation length. We will also suppose that
each aggregate has the same number m of RBC. The
ultrasound wave is scattered by these aggregates
which therefore will be called the scattering
units (S.U.).
If V is the number of S.U. per unit volume and
V|^ the RBC volume, the usual hematocrit number H
is given by:
H = mv,\) n)
165
However we are interested in the volume fraction
of the S.U. and not the static hematocrit H. We
will introduce a "dynamic" hematocrit number H'
as follows:
H' = vv'
su
where viy is the excluded volume of the S.U.
(fig. 1) which takes into account the S.U. volume,
vsu. and the trapped plasma between the S.U.:
da = — dfi
i
In back-scattering experiments, the angles 6
and (j) values are 180° and 90° respectively (fig.
2). The detecting solid angle depends on the
probe surface S and the distance x between the
probe and the S.U. One can take into account
these parameters by introducing a geometrical
factor g(x) = S/x^.
su
su
where p^^ is a dimensionless number greater than
or equal to 1. The volume of one S.U. must also
take into account the trapped plasma inside the
S.U.:
su
"^^hPh
where p^ is another dimensionless quantity great-
er than or equal to 1. If the RBC were perfectly
packed without free volume for the plasma, pj^
and ph would be equal to 1. For an imperfect
packing, they are greater than 1. The dynamic
hematocrit H' is greater than H because in the
course of their motions the RBC's and S.U.'s car-
ry with them a fraction of the plasma. Its value
is:
H'
'hPhPh
Using this model, the S.U. radius is given by:
A =
4^ "^^Ph
ll3
(2)
(3)
Trapped plasma
Inside the S.U.
one S.U.
' Trapped
, ' ' plasma
between
^the S.U.
Fig. 1. Blood model used, showing the transient
spherical aggregation (scattering unit
S.U.).
Ultrasound Back-Scattering by One S.U.
Let io be the incident wave intensity per unit
area of a spherical scattering center, with radi-
us A smaller than the wavelength. The scattered
intensity per unit solid angle in the direction e
is i[). The detecting solid angle being du, the
differential scattering cross-section is:
Emetting probe
Receiving probe
\
Fig. 2. Scattering coordinates.
We can only measure intensities emitted and
received by the probe, Iq and Iq. They differ
from io and ip because of the attenuation of the
incident and scattered beams by the medium be-
tween the probe and the S.U. They also differ
because the ultrasonic field varies with distance.
We can lump all these factors in a general func-
tion of the distance x, G(x) so that one can
write
(4)
The factor G(x) is not essential in our final
results. It could be determined by a substitu-
tion method as described, for instance, by Sigel-
mann and Reid [3].
The differential scattering cross-section is
known for some simple models [7]. Using the re-
sults for an elastic sphere, the back-scattered
intensity is finally:
I,G(x)
w'^A^
9c^
V C2p
+ 3
2p + P.
(5)
where c and Cg are the sound wave velocities in
the S.U. and the plasma respectively, and p and
Pp the corresponding densities. We assume that
the S.U. coefficients, c and p, are identical to
the RBC coefficients, cp, and p^.
Back-Scattering by Whole Blood
The scattering of dense particle medium is a
complex theoretical problem which has already
been treated by Twersky [8-10] and applied by
Shung, Sigelmann and Reid [4,5]. The following
simplified approach can be used. Let s' = qjuS
be the functional diametral area of one S.U.,
166
where s is the diametral area of one S.U. and
qsu is a dimensionless number, usually greater
than 1, and which is characteristic of the S.U.
packing. For a spherical S.U. one has:
s' = qsu'fA^.
We are interested by the intensity I^x scatter-
ed by a cylindrical volume element having the
same axis as the ultrasonic wave with height Ax
and base area S at a distance x from the probe
(fig. 3).
Probe
If 1
cylindrical tube
tilled with blood
Fig. 3. Simplified diagram of the sedimentation
tube with the echographic probe.
We will consider in this volume an elementary
slab (S, 4A) defined by the two sections with
distances (x + 2A) and (x - 2A). The maximum
number of S.U. which can exist in this slab is
2N = 2S/s'; we will call N the number of sites,
each composed of two cells, which can be occupied
by 0 or 1 or 2 S.U.
Figure 4 shows a simplified slab with 7 sites
occupied in such a way that the ensemble average
of the hematocrit is 4/7. Eq. (5) only applies
to the sites "f"; for the other ones, eq. (5) in-
dicates that the scattered intensity is zero.
■2A
X +2A
Fig. 4.
Statistical model. The slab (x - 2A,
x + 2A) has 7 sites with an occupation
probability of 4/7. Only the "f" emplace-
ments have a favorable configuration for
backscattering.
By a simple calculation in order to find the
probability that a given site is in a favorable
configuration (as regards back-scattering), it can
easily be shown that the number of sites, Nf, which
have a favorable configuration in an elementary
slab 2A is:
NH' (1 - H' )
(6)
Assuming that Iq and Iq do not depend on x in
the volume Ax (this point is justified by the low
values of the attenuation and reflexion coeffi-
cients in blood), each slab 2A back-scatters with
the intensity:
I^^ = NH'(1 - H')Ip
(7)
As there are Ax/2A elementary slabs in the in-
terrogated volume (Ax, S), one has:
Iax
Ax
^(1
2A
H')I
D
This expression can be further transformed by
replacing N by S/s', and H' by PsuPhH. using eq.
(5) for I[). As we have used an impenetrable
spheres model with a cubic packing, the factor
Psu/Psu is equal to 3/2. Finally one obtains:
Ax
G(x)Kmv^p2H(l - P^Psu")
(8)
In this derivation we have added the individual
intensities and not the wave amplitudes, thus
ignoring the interference problem. This is justi-
fied in a first approach because the distribution
of the scatterers in the interrogated volume is
random.
Back-Scattering Fluctuations
Examination of eq. (8) shows that the only
fluctuating quantities are H or H' (the packing
factors do not fluctuate as we have assumed a
spherical model for the aggregates. It is also
assumed in the model that m is constant. For an
elementary variation dH' of H' in a slab (2A), the
scattered intensity will vary according to eq.
(7):
^^2A " ^d"^'^ " 2H' )dH'
The relative mean quadratic fluctuation F2A
is defined as follows:
<(dl2A)'>^'
2A
<I2A>
and can be written:
'^2A <H' (1 - H' )> ^
2A
a<(dH' )2>i/2
167
If one transforms dH' into dv, one obtains:
As in the initial model the site occupancy proba-
bilities are independent, the S.U. dis,tribution
obeys Boltzmann statistics and <(dv)2>^ is given
by [11]:
<(dv)2>i/2 =
(2AS)i'2
Therefore F2/\ can be written:
= a(2ASv^p^p^^m)V2<H.>i/2
The corresponding fluctuation coefficient for
the blood column Ax is F^^:
<(dl^^
Assuming independent fluctuations for the
Ax/2A elementary slabs in Ax, the total fluctua-
tion is given by:
to = 2 IT 5.2 10^ s'l
Po = 1.078; p = 1.223
Co = 1550 m s-i; c = 1610 m s"!
G(x) = 1 (negligible attenuation)
m =2 (slight aggregation)
— theoretical curve
, SHUNG et al results
10 M Ml 40 50 60 H( percent)
'^.-C^Wlu^'y (9)
<H'>ife<(l - 2H')2>ig
<H'(1 - H')> Ax^
This relation shows that the standard deviation
of the fluctuations, given by F^^ *^ax^' ''^ P'"'^"
portional to m''/^.Ax''2.
Discussion
In our model, the interactions between RBC
lead to monodisperse spherical aggregates. In-
deed, microscopic observations show a '^eversible
aggregation in normal and pathological blood [12].
Our picture may be oversimplified, as these
transient aggregates are polydispersed in dimen-
sion and shape. We feel however that introducing
polydispersity function would not change our
treatment fundamentally.
Eguation (8) is very similar to a previous re-
sult obtained by Shung et al. [4] using Twersky's
theory [8-10]. However, as this theory predicts
that the back-scattering goes through a maximum
for H = 0.50, Shung et al . introduced an empiri-
cal fitting constant as their experimental re-
sults showed a maximum for hematocrit values be-
tween 0.25 and 0.30
We have plotted on figure 5 their experimental
results and a theoretical curve obtained by using
eq. (8) and the following values:
Fig. 5. Back scattering of whole blood as a func-
tion of hematocrit: correlation between
the experimental results of Shung et al .
[4,5] and theoretical values calculated
according to eq. (8).
Shung et al . ' s scattering coefficient a is iden-
tical to:
I
SAx
The quantitative agreement between the two re-
sults seems satisfactory, our treatment can di-
rectly explain the experimental results of Shung
et al. [4,5].
The aggregation dependence of the scattered
intensity by whole blood is also shown in eq. (8)
and has recently been demonstrated by Shung and
Reid [13].
In order to derive the fluctuation coefficient,
it has been assumed that the probability of find-
ing a S.U. in a given elementary volume is propor-
tional to the S.U. number concentration and in-
dependent of the occupation state of the neighbour-
ing elementary volumes. This would be the case
for pure Brownian motion. However, as has already
been pointed out by Atkinson and Berry [6], the
thermal diffusion of RBC's is much too slow (10^ s)
to travel one wavelength.
Two other mechanisms could be postulated. The
first one is convection streams which break apart,
associate, rotate and translate the aggregates,
so that the occupancy of a given elementary volume |
fluctuates to and fro in the occupation probability.
The second explanation can be found in the disso-
ciation-association kinetic of the S.U. In that |
case it is m which would fluctuate. Further ex- I
168
periments are needed in order to assess the respec-
tive importance of the mechanisms.
Summary
The ultrasound back-scattering of blood has been
shown to be increased when the erythrocyte sedi-
mentation rate is high (fig. 1). Moreover, in this
case, temporal fluctuations of the back-scattered
intensity have also been demonstrated. In order
to explain these results, a simplified theory of
ultrasound back-scattering by blood is presented.
The following model of the blood structure is
used: the erythrocytes are associated to form
spherical transient aggregates having the same cell
number, m. The acoustic wave is scattered by these
clusters called "scattering units".
The scattered intensity is calculated for one
kinetic unit taking a spherical elastic model,
then for a blood unit volume, when the hematocrit
H is very low. It is found that the back-scattered
intensity is proportional to H and to m.
When H is not negligible, a simplified statis-
tical theory is proposed. The main result is that
the back-scattered intensity is still proportional
to m but that it passes through a maximum value
when H gets near 0.3.
An explicit relation between the mean relative
value of the intensity fluctuation, H and m is
given.
Therefore, the aggregation state of blood can
in principle be determined through ultrasound scat-
tering studies. Further theoretical developments
are in progress so as to take into account the dis-
tribution of the number of erythrocytes in each
kinetic unit and the variation of this number with
H.
[8] Twersky, V., On scattering of waves by random
distribution: I. Free space scatterer
formalism, J. Math. Phys. 3^, 700 (1962).
[9] Twersky, V., On scattering of waves by random
distribution: II. Two space scatterer
formalism, J. Math. Phys. 3, 724 (1962).
[10] Twersky, V., Acoustic bulk parameters of
random volrame distributions of small scat-
terers, J. Acoust. Soc. Am. 36, 1314 (1964).
[11] Landau-Lifchitz, Statistical Physics (Mir
ed., Moscow, 1967")^
[12] Schmid-Schonbein, M. , Gallasch, G., Gosen,
J. v., Volger, E. , and Klose, M. J., Red
cell aggregation in blood flow, Klin. Wschr.
54, 149 (1976).
[13] Shung, K. K. and Reid, J. M., Ultrasonic
detection of erythrocyte aggregation, 29th
ACEMB Sheraton-Boston, Boston, Massachusetts,
6-10 November 1976.
References
[1] Hanss, M. , Boynard, M. , and Perrin, J.
Erythrocyte sedimentation by an echographic
method, Biomedicine 25 (3), 81 (1976).
[2] Reid, J. M., Sigelmann, R. A., Nasser, M. G.,
and Baker, D. W. , The Scattering of Ultrasound
by Human Blood. Proceedings of the 8th Inter-
national Conference in Medicine and Biological
Engineering (1969).
'j [3] Sigelmann, R. A. and Reid, J. M. Analysis
and measurement of ultra-sound backscatter-
ing from an ensemble of scatterers excited
by sine wave bursts, J. Acoust. Soc. Am. 53,
1351-1355 (1973).
[4] Shung, K. K. , Sigelmann, R. A., and Reid,
J. M. , The scattering of ultrasound by red
blood cells, Applied Radiology 77 (Jan.,
Feb. 1976).
[5] Shung, K. P., Sigelmann, R. A., and Reid,
J. M., The scattering of ultrasound by
blood, I.E.E.E. trans, on biomed. Enain.
BME 23 (6), 460 (1976). ^
[6] Atkinson, P. and Berry, M. V., Random noise
in ultrasonic echoes diffracted by blood,
J. Phys. Math. Nucl. 7 (11), 1293-1302 (1974).
[7] Morse, P. M. and Ingard, K. U., Theoretical
Acoustics (McGraw Hill, New York, 1968).
169
CHAPTER 6
TUMOR DOPPLER SIGNATURES
171
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed., National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979)
TUMOUR DETECTION BY ULTRASONIC DOPPLER BLOOD-FLOW SIGNALS
P. N. T. Wells, M. Halliwell, R. A. Mountford,
R. Skidmore, A. J. Webb and J. P. Woodcock
Bristol General Hospital and Bristol Royal Infirmary
Bristol, United Kingdom
An in situ cancer is avascular and harmless. Vascul ari sation is necessary before
a tumour is potentially capable of rapid growth. Distinctive continuous-wave ultrasonic
Doppler signals have been obtained transcutaneously from malignant breast tumours.
Similar signals have been detected from abdominal malignancy, by means of a pulsed
Doppler. These signals may form the basis of a practicable method of screening for
breast cancer, and of improving the accuracy of cancer diagnosis.
Key words: Blood flow; breast cancer; diagnosis; Doppler; screening; ultrasonics.
1. Tumour Blood Flow
A single aberrant cell is almost certainly the
origin of every solid malignant tumour. This
cell divides to form a colony of malignant cells,
called an i_n situ cancer. An ijT_ situ cancer is
avascular and harmless, and cannot grow beyond
the volume limited by diffusion of nutrients and
wastes. The event which converts an in_ si tu
cancer into a "tumour" is apparently the release
by the malignant cells of a diffusable chemical
substance, called the "tumour angiogenesis fac-
tor" [1]^. Once the tumour is vascul ari sed it is
potentially capable of rapid growth. The blood
flow in normal tissue like that of the breast is
in the range 10 to 30 ml min-i kg-^ [2]. The blood
flow in lymphomas is about 380 ml min~^ kg"^, in
differentiated tumours, 140 ml min"^ kg"^, and in
anaplastic tumours, 110 ml min"^ kg"i [3].
The possibility that changes in blood flow
might be used to detect malignant tumours is dis-
cussed in this paper, and some preliminary results
are presented. Most of the experimental work was
done with breast lesions, because of their acces-
sibility and the potential importance of the method
in screening for breast cancer.
The blood vessels which serve the breast are il-
lustrated in figure 1. The mass of a typical fe-
male breast is about 0.3 kg. Assuming that the
three main arteries each serve one third of the
breast, the normal flow rate in each artery would
be around 1 to 4 ml min"^. A malignant tumour with
diameter of 20 mm would require a flow rate of
about 0.5 to 1.5 ml min*^. Therefore, in an in-
dividual, there might be a significant difference
between the flow in an artery supplying a tumour
and part of the breast, and the flow in the cor-
responding artery serving the other normal breast.
There is at present no non-invasive method of mea-
suring blood flow volume which would enable this
^Figures in brackets indicate literature
references at the end of this paper.
Fig. 1. Principal blood vessels serving the right
breast.
to be tested. It may be that a change might occur
in the pulsatility index of the flow [4], but this
would be difficult to establish since every pair of
arteries (serving the two breasts) would need to
be studied.
In biomedicine, liquid flow through small ducts
has previously been recognised in backscattered
ultrasonic Doppler frequency-shifted signals [5]
in two situations. The first of these is in the
placenta [6] and the second, in the fetal lung dur-
ing breathing [7]. Recently it has been reported
that characteristic blood flow signals are asso-
ciated with neovascular flow in malignant tumours
[8j and additional data are presented in this paper.
2. Breast Investigations
Breasts were examined as illustrated in figure
2. A regular continuous-wave ultrasonic Doppler
blood-flow detector [5] was used. The hand-held
probe had a diameter of about 10 mm, and contained
two separate 8 MHz transducers, one for transmit-
ting and one for receiving. The transducers were
173
HJNO-Hilll HKOBl
IRANSMIIIINC
IfiiHSOUCtB
Fig. 2. Method of obtaining ultrasonic Doppler
blood-flow signals from the breast.
(a)
(kHz)
(b)
fREOUEHC?
Ikm)
FREOUEdCt
IkHi]
aligned so that the volume sensitive to motion ex-
tended axial ly beyond the end of the probe. For
each patient, the Doppler signals obtained from
over the lesion were compared with those from the
corresponding site on the opposite breast. Both
breasts of each normal individual were thoroughly
explored to search for Doppler signals which were
asymmetrical and characteristic of neither normal
arterial nor venous flow.
The results are summarised in table 1. Sono-
grams of the signals recorded from a normal mam-
mary artery, a malignant breast tumour, and tis-
sues at the corresponding site in a normal breast,
are shown in figure 3.
Table 1. Ultrasonic Doppler signals obtained from
breast in various clinical conditions.
Confirmed diagnosis^
Number of patients with
neovascularisation signals
None
Weak
Strong
Normal
2
0
0
Mastitis
1
0
0
Mammary duct
pa pi 1 lomatosis
1
0
0
Mammary duct
ectasia
2
0
0
Benign mammary
dysplasia
2
1
0
Cyst
1
2
0
Fi broadenoma
0
1
0
Lymphoma
0
0
1
Carcinoma
0
0
2
Number of patients = 16.
3. Investigations of Other Tumours
Signals apparently associated with malignant
neovascularisation have been detected in the ab-
domen. Two-dimensional scanning was used to locate
a pancreatic tumour. The ultrasonic beam of the
scanner probe was directed through the tumour.
The probe was then connected to a 2 MHz pulsed
Doppler [5]. Blood flow signals were detected when
the Doppler was range-gated into the tumour.
Fig. 3. Sonograms of ultrasonic Doppler blood-flow
signals obtained from the breast, (a)
Normal individual: branch of internal
mammary artery; (b) malignant tumour; and
(c) flow at site corresponding to (b) in
other breast of the same patient.
4. Discussion
A. Breast Cancer
Breast cancer is the most common malignancy in
Western women. In the United Kingdom, breast
cancer affects at least 80 in 100,000 females;
it kills 1 in 50 women, 1 in 75 women between the
ages of 35 and 60 years, and 1 in 3,500 females
of all ages every year. Breast cancer is the
main cause of all deaths among women between the
ages of 40 and 44. Extrapolating from 1973 U.S.
data [9], the present economic cost of breast
cancer in the United Kingdom is in the order of
$40 million per year (and $200 million per year
in the United States).
Earlier diagnosis and treatment of breast cancer
might improve the survival time of the patient
[10]. If the cancer is removed whilst it is still
in situ, the patient is cured. The next step in
the natural progress of the disease is when the
growing cancer begins to invade the surrounding
tissue. For a time, the cancer cells remain local-
ised. If diagnosed and treated at this stage, the
5-year survival rate is around 85 percent. If un-
treated, the cancer metastases, with regional in-
volvement of the lymph nodes which drain the breast,
and the 5-year survival rate falls to about 53 per-
cent. If left untreated, metastases occur in more
distant parts of the body. This advanced cancer is
virtually incurable, although the time taken for
the patient to die is variable.
The treatment of patients with early breast
cancer is more successful than that of those in
whom the disease is advanced. Breast cancer is
usually discovered when the patient herself feels
a lump. Even in a "suitable" breast, the smallest
lump which can be detected by manual palpation is
not much less than 10 mm in diameter [11]. The
risk of the presence of distant metastases from a
tumour with a doubling time of 1.5 months increases
from 22 percent when its diameter is 1 mm, to 43
percent, when it is 10 mm. Once a lump has been
discovered in a breast, present diagnostic proced-
ures are already adequate, although there may be
174
disagreement about the best course of treatment if
the lump is a malignant tumour. The possibility
exists, howt.^r, that diagnosis of pre-symptomatic
lesions by an effective screening procedure might
lead to a reduction in breast cancer mortality as
a result of earlier treatment.
Mammography, either conventional or xeroradio-
graphic, seems to be the best contemporary method
of detecting early breast cancer [12]. Unfortunate-
ly, however, mammography as a screening procedure
confers no benefit on women who are well and under
the age of 50 [13]. In these women only 19 percent
of cancers would not have been detected without
mammography, and so the benefit (which has to be
set against the risk of radiation-induced cancer
as a consequence of mamography [14]) is small.
Even if technical developments were to result in a
reduction in x-ray exposure to an acceptable level,
the logistics of interpreting the vast number of
mammograms which would be obtained in a screening
programme pose apparently insoluble problems of
manpower, boredom and expense.
Thermography has been used to study breast
tumours [15] but initial enthusiasm for the method
has declined. Thus, in a series [16] of 2523
volunteers, 1 patient out of 4 who developed cancer
within 18 months of the examination was detected by
thermography; 344 had abnormal or suspicious scans,
but no abnormality. These and other results are so
poor that the method (at least when used alone)
seems to have no role in screening for breast
cancer.
Conventional two-dimensional pulse-echo ultra-
sonography is capable of producing images of the
breast of quite high resolution [17-20]. Un-
fortunately, the normal breast is a complex ir-
regular structure, the recognition without prior
knowledge of small tumours is very unreliable, and
there seems to be no way of automating the analysis
of the vast number of scans which would result from
even only a modest screening programme. The re-
sults reported in the present paper indicate that
characteristic Doppler signals may be a cancer-
specific ultrasonic tissue signature. They might
form the basis of a practicable breast cancer
screening method.
B. Other Tumours
Cancer-specific Doppler signals could have an
important place in the interpretation of the in-
adequate data which sometimes results from con-
ventional diagnostic tests such as radiography,
echography, scintigraphy, and computerised tomo-
graphy. These opportunities, which are quite
distinct from screening, include, for example,
the differentiation of small solid and cystic
renal lesions, and the differentiation of
ophthalmic tumours and organised haematomas.
References
[1] Folkman, J. and Cotran, R. , Relation of
Vascular Proliferation to Tumor Growth, in
International Review of Experimental Pathology,
C. W. Richter and N. A. Epstein, eds.. Vol. 16,
pp. 207-48 (Academic Press, New York, 1976).
[2] Woodcock, J. P., Theory and Practice of Blood
Flow Measurement (Butterworths , London, 1975).
[3] Mantyla, M. , Kuikka, J., and Rekonen, A.,
Regional blood flow in human tumours with
special reference to the effect of radio-
therapy, Br. J. Radiol. 49, 335-8 (1976).
[4] Woodcock, J. P., The Significance of Changes
in the Velocity/Time Waveform in Occlusive
Arterial Disease of the Leg in Ultrasonics in
Medicine, L. Pi 1 i pczyiiski , ed., pp. 243-50
(Polish Scientific Publishers, Warsaw, 1970).
[5] Wells, P. N. T. , Biomedical Ultrasonics
(Academic Press, London, 1977).
[6] Hunt, K. M. , Placental localization using the
Doptone foetal pulse detector, J. Obstet.
Gynaec. Br. Commonw. 76^, 144-7 (1969).
[7] Boyce, E. S., Dawes, G. S., Gough, J. D., and
Poore, E. D., Doppler ultrasound method for
detecting human fetal breathing in utero, Br.
Med. J. 2, 17-8 (1976).
[8] Wells, P. N. T., Halliwell, M., Skidmore, R.,
Webb, A. J., and Woodcock, J. P., Tumour
detection by ultrasonic Doppler blood-flow
signals. Ultrasonics (to be published).
[9] Scitovsky, A. A. and McCall, N., Economic
Impact of Breast Cancer in Frontiers of Radia-
tion Therapy and Oncology, J. M. Vaeth, ed..
Vol. 11, pp. 90-101 (Karger, Basel, 1976).
[10] Brinkley, D. and Haybittle, J. L., A 15-year
follow-up study of patients treated for
carcinoma of the breast, Br. J. Radiol. 41
215-21 (1968).
[11] Slack, N. H., Blumenson, L. E., and Bross,
I. D. J., Therapeutic implications from a
mathematical model characterizing the course
of breast cancer. Cancer, N.Y. 24, 960-71
(1969).
[12] Feig, S. A., Shaker, G. S., Schwartz, G. F. ,
Patchefsky, A., Libshitz, H. I., Edeiken, J.,
Nerlinger, R. , Curley, R. F., and Wallace,
J. D. , Thermography, mammography and clinical
examination in breast cancer screening.
Radiology 122. 123-7 (1977).
[13] Strax, P., Venet, L., and Shapiro, S., Value
. of mammography in reduction of mortality from
breast cancer in mass screening. Am. J.
Roentg. 117, 686-9 (1973).
[14] , Radiation-Induced Breast Cancer,
Br. Med. J. I, 191-2 (1977).
[15] Williams, K. L., Lloyd Williams, F. J., and
Handley, R. S. , Infra-red thermometers in
the diagnosis of breast disease. Lancet 2,
1378-81 (1961).
[16] Hitchcock, C. R., Hickok, D. F. , Soucheray,
J., Moulton, T. , and Baker, R. C. , Thermo-
graphy in mass screening for occult breast
disease, J. Am. Med. Ass. 204, 419-22 (1968).
175
[17] Wells, P. N. T. and Evans, K. T. , An immer-
sion scanner for two-dimensional ultrasonic
examination of the human breast. Ultrasonics
6, 220-8 (1968).
[18] Jellins, J., Kossoff, G., Reeve, T. S., and
Barraclough, B. H., Ultrasonic gray scale
visualization of breast disease, Ul trasound
Med. Biol. U 393-404 (1975).
[19] Baum, G., Ultrasound mammography. Radiology
122, 199-205 (1977).
[20] Kobayaski, T. , Gray-scale echography for
breast disease. Radiology 122, 207-14 (1977).
176
CHAPTER 7
PROPAGATION THROUGH BONE AND SKULL
177
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer , ed.. National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
A THEORY RELATING SONIC VELOCITY TO MINERAL CONTENT IN BONE
Sidney Lees
Forsyth Dental Center
Boston, Massachusetts 02115, U.S.A.
and
Carel L. Davidson
University of Amsterdam
Amsterdam, The Netherlands
Bony tissues consist primarily of mineral hydroxyapati te (HAP) crystallites em-
bedded in a matrix of a much softer material, collagen. Currey and others suggested
that bone is a two-phase composite, like mineral filled plastics, but the known laws
of mixtures give at best a crude approximation of the observed elastic properties of
bone. We have investigated the problem by measuring the ultrasonic velocity in min-
eral filled particulate composites as a function of the mineral concentration.
It was found that the ultrasonic velocity can be predicted for some mineral filled
plastics by applying the Reuss formalism to the longitudinal elastic modulus, indicat-
ing that these are families of Reuss solids in some sense. Other mineral filled
plastics do not seem to obey this rule, the ultrasonic velocity being greater than
predicted. Bone appears to belong to this latter class of particulate composites.
The literature indicates that in certain situations the mineral filler affects
the plastic matrix, giving the plastic constituent a higher modulus. A maximum
longitudinal modulus is ultimately attained so that two Reuss formalism bounds can
be obtained, the lower one calculated from the modulus of the unfilled plastic, the
upper using the maximum modulus. The ultrasonic velocities of a system of fluorapa-
tite (FAR) filled epoxy was found to lie between such bounds.
Currey and his successors assumed bone collagen has the same elastic properties
as tendon collagen and that these are invariant with respect to contained HAP. The
literature shows the contrary, that bone collagen, even when demineral ized , is more
highly cross linked than any other collagen. Moreover, the literature indicates that
HAP crystallites are chemically bonded to the collagen molecules.
It is postulated that bone collagen is stiffened because HAP crystallites form on
the intermolecular cross links, encasing them and making their effective lengths very
short. It is shown that the sonic velocity for bone can be bounded by two Reuss
formalism curves in the same manner as for FAP epoxy.
Keywords: Bone; collagen; crosslinking modification; curve! i nki ng ; hydroxyapatite;
sonic velocity.
1. Bony Tissues
Bony tissues are essentially a mineral,
hydroxyapatite (HAP), embedded in an organic ma-
trix, mostly collagen. HAP is an hexagonal
crystallite which in bone and dentin is usually
less than 100 nm in any dimension. The shape
and size distribution of the crystallite in bone
has been studied for many years with inconclusive
results, except that they are small. Certain x-
ray diffraction studies indicate the crystallites
are needlelike, others that they are platelets.
Electron micrographs show platelets, but the
needle form is not thereby excluded. There is
reason to believe both forms are present de-
pending on the site of formation. There is also
a strong contention that a significant fraction
of the mineral is amorphous [1]^.
Collagen is a generic term for a class of
protein that make up much of the body tissue.
Bone collagen differs from other collagens in
the body because it is so insoluble, indicating
a high degree of intermolecular cross linking.
The organic part of bone constitutes about 65
percent by volume of the tissue of which 95 per-
cent is collagen. The mineral occupies 35 per-
cent of the volume. Since collagen is such an
important body constituent, its chemistry has
been studied intensively for many years and is
^Figures in brackets indicate literature
references at the end of this paper.
179
still a major field of biochemistry. Very re-
cently the sonic and elastic properties of some
types of collagen have begun to be studied be-
cause of the need to determine the presence of
collagen in tissues and to identify its extent.
All collagens are highly structured materials
in a multilevel hierarchal order. The molecular
weight of the basic collagen molecule, defined
as tropocollagen, is approximately 300,000
daltons, which is a very large molecule even in
organic chemistry. The molecule is much like a
piece of spaghetti having a wet diameter of
1.5 mm and a length of about 300 nm. It is dif-
ficult to draw to scale and all figures in this
paper are purely schematic. As noted, bone col-
lagen differs from other body tissue collagens
by an extensive network of intermolecular cross
links that render it insoluble in even the most
potent of solvents. On the other hand, tendon
collagen is reported to be mostly lacking in
intermolecular bonds but having many hydrogen
bonded intramolecular links, so that it can be
dissolved and reconstituted readily. While other
tissues like arteries and joints calcify, there
is reason to think that the special structural
characteristics of bone collagen cause it to
calcify in a unique manner [2].
2. Two Phase Mi neral -Fi 1 led
Plastic Composites
It is the contention of this paper that the
mechanical properties of bone can only be under-
stood in the light of the chemical interrelation-
ship between the mineral crystallites and bone
collagen. Previously Currey [3], Welch [4] and
Katz [5] have considered bone to be a mechani-
cally mixed two phase mineral-filled polymer
composite. Their concept yields a crude rep-
resentation of the variation of elastic modulus
with mineral content, but the detailed correla-
tion is not good. Currey [6] even suggested
that the crystallites are much longer than ob-
served in order to attribute a fibrous mineral
structure to bone to account for the observed
elastic and strength properties.
We believe that the difficulties of our pred-
ecessors are due to several causes. The elastic
modulus of collagen is not well known and there
has been no distinction among the various types.
Previous investigators used estimates for tendon
fibers rather than bone collagen. Except for
Mason's [7] sonic velocity measurement on kan-
garoo tail tendon fibers, the estimates of the
elastic modulus were based on low strain rate
tests. Secondly, there was no attempt to examine
the chemical structure of collagen or the effect
of the structure on the properties of bone. Col-
lagen was considered a structureless homogeneous
isotropic continuum and the short range order of
body tissue collagen was ignored. Thirdly,
there is little understanding even at this date
how the HAP crystallites are laid down despite a
long and intensive investigation over many years
by many investigators. Fourth, the estimates of
the elastic modulus of bone have been mostly
those obtained by standard stress-strain testers.
Bone is vi scoelasti c . Strain rates and the magni-
tude of the strain in conventional stress-strain
testing causes hysteresis and often permanent
distortion. In this paper we use only ultrasonic
velocity measurements where the amplitude of dis-
placement is of the order of an angstrom and the
period is much less than any relaxation time of
the medium.
3. Ultrasonic Measurements
It was the availability of measurements of the
ultrasonic velocity of several hard tissues that
led to the elastic theory presented here. Con-
ventional procedures for measuring elastic prop-
erties are difficult because bone is a visco-
elastic material and the data must include the
rate of change of strain as well as the strain
itself. Moreover, the sample is significantly
affected by the test process, so that repeated
tests are not consistent. Such a situation re-
quires further interpretation to make sense of
the data which is particularly difficult where
the material (bone) varies widely due to its
biological origin. Ultrasonic velocity measure-
ments, on the other hand, can be repeatedly per-
formed on the same sample with an uncertainty
that depends mostly on the measurement technique
and not on the sample. The contribution of
biological variation can be treated separately
apart from the measurement process.
There is a need to relate the elastic prop-
erties obtained at high strain rates with the
more conventional techniques. Papadakis [8] has
designed and tested an infrasonic resonator for
testing plastics at about one hertz but which
he says can be increased to 1000 hertz. It will
be useful to find the sonic velocity measurements
in bone at these low frequencies, but the limita-
tions imposed by the inhomogeneities of biologi-
cal material will require considerable modifica-
tion of the Papadakis equipment.
4. Rule of Mixtures
It is reasonable to consider bone to be a two-
phase mineral-filled composite as Currey did,
but it is necessary to know the relationship by
which the elastic properties of the composite
can be calculated from those of the constituents.
A number of rules of mixtures have been proposed
for estimating the elastic moduli of the compos-
ite but each has limited applicability. General-
ly they work reasonably well when the properties
of the two constituents are nearly alike but
when there is a tenfold or greater ratio in
density, the mixing rules do not seem to apply.
A number of mineral filled plastics were
studied to determine a relationship between
longitudinal sonic velocity and mineral con-
tent. The first series included tungsten-
filled vinyl, crystabolite-filled polymethyl-
methacrylate, and crystabolite-filled epoxy.
It was discovered that one of the known mix-
ing rules did apply. Reuss postulated an ex-
pression of the form:
1 _ Vj_ V2.
K " Ki K2
(1)
where
^1 '
volume fraction of the plastic
consti tuent
volume fraction of the mineral
constituent = 1 - Vj
elastic modulus of composite
elastic moduli of constituents.
180
In this application the longitudinal modulus
was used rather than the bulk and shear moduli
which was Reuss's intention. Since
K
k + 4/3
(2)
where k = bulk modulus and u = shear modulus,
another value for the longitudinal modulus can
be calculated from Reuss's expression applied
first to the bulk modulus and then to the shear
modulus, but it will be quite different from the
value yielded by eq. (1) for the longitudinal
modulus directly inserted. Consequently, the
expression in eq. (1) is designated the Reuss
formal i sm.
The sonic velocity is calculated as usual
from the relation:
c = K/p (3)
where p = Vjpj + P2V2 and p = density.
Figures 1, 2 and 3 taken from Lees and David-
son [9] show how well Reuss's formalism applies
to these examples. It is not obvious why and we
do not have an explanation. It is characteristic
of these figures that the sonic velocity de-
creased, or at best does not increase much, as
the mineral content increases on the left hand
side of the curve, even though calculations show
the longitudinal modulus of the composite is in-
creasing. It was found that the density in this
regime increases much more rapidly than the
elastic modulus. When Vj is large and K2 is ten
times Kj the first term on the right side of eq.
(1 ) dominates.
1-
VOIGT
REUSS
+ EXPERIMENTAL VALUE
0 10 20 30 40 50 60 70 80 90 100
VOLUME % CRYSTABOLITE
Fig. 2. Polymethylmethacrylate-crystabol ite
system (from Lees and Davidson [9]).
6
0 H 1 1 1 1 1 1 1 1 1 —
0 10 20 30 40 50 60 70 80 90 100
VOLUME, PERCENT TUNGSTEN
Fig. 1. Voigt-Ruess bounds for vinyl -tungsten
composite system with experimentally
determined curve (from Lees and
Davidson [9]).
+ EXPERIMENTAL VALUE
1 -
0 -I 1 1 1 1 1 1 1 1 1
0 10 20 30 40 50 60 70 80 90 100
VOLUME, PERCENT CRYSTABOLITE
Fig. 3. Epoxy-crystabol i te composite system
(from Lees and Davidson [9]).
181
5. Modified Conditions for Rule of Mixtures
As the investigation proceeded a new kind of
mineral-filled plastic, f 1 uorapati te-f i 1 1 ed
epoxy, was found which apparently does not fol-
low Reuss's formalism. Fluorapatite (FAP)
closely resembles HAP but unlike HAP it can be
obtained in large crystals. Powdered FAP is
much more crystalline and well behaved crystal lo-
graphically than HAP. Both FAP and HAP are
surface active materials. In figure 4, the
lower curve shows the variation of sonic veloci-
ty of the composite with mineral content when
the longitudinal modulus of the unfilled plastic
is taken as the value of Kj . It may be observed
that all the measured values are above the lower
curve and that there is some scatter among them.
Q J I I 1 I I I I I I
7-
LIMITING
SONIC VELOCITY
BOUNDARY
1 -
0-1 1 1 1 1 1 1 1 1 1
0 10 20 30 40 50 60 70 80 90 100
VOLUME, PERCENT FLUORAPATITE
Fig. 4. Epoxy-f 1 uorapatite composite system
(from Lees and Davidson [9]).
The only difference between the situations of
figures 3 and 4 is the replacement of one mineral
filler by another. It must be concluded that
sometimes the filler can influence the mechanical
behavior of the plastic matrix. Now it is well
known that polymers become stiffer as the cross
linking density increases so it is inferred that
FAP must cause the epoxy to increase the cross
linking density. A similar situation was de-
scribed by Kumins [10] who found that titanium
dioxide causes an increase in cross linking
density in epoxy, apparently by a factor of ten
when about 3 volume percent titanium dioxide was
added. Burhans et al . [11] showed a strong de-
pendence of the elastic moduli on the maximum
cross link length, so that either an increased
cross linking density, or a decreased maximum
cross link length, or both can increase the
elastic modul i .
In order to see how this could be applied to
the FAP-epoxy composite system, the maximum
value for the compressive modulus reported by
Burhans et al . was used as a basis for estimat-
ing the corresponding maximum value for the
longitudinal modulus. A new Reuss curve was
calculated using the higher modulus. When plot-
ted on figure 4 it provided an upper bound, so
that all the experimentally determined velocities
fell between two Reuss curves. This is inter-
preted to mean that FAP can affect the cross
linking density of the matrix. The scattered
values of the ultrasonic velocity indicates a
variability in the cross linking density from
one experiment to the next. The samples were not
always prepared the same way and the changed
conditions could possibly modify the influence
of the filler. For example, the samples were
compressed, after mixing, in an attempt to
achieve higher mineral content. This caused a
gray exudate to ooze from the mold, the gray
color being attributed to the very fine FAP
particles. But these fine particles are most
likely to have maximum influence on cross link-
ing density because of their increased surface
area .
6. Structure of Collagen
It is now possible to study ultrasonic wave
propagation in bone with these two concepts in
mind. First, that for a specific two phase
mineral-filled polymer the longitudinal modulus
of the composite can be found from the Reuss
formalism and, second, the longitudinal modulus
of the matrix can be affected by the presence
of the f i Her.
While collagen is a polymer and bone may be
considered a two-phase mineral-filled polymer,
bone is not formed by starting with a mix of
mineral and monomer. Collagen is laid down
first and then it becomes mineralized when the
HAP crystallites are deposited. When bone is
demineral ized the collagen matrix can be re-
covered as a rubbery solid having the form and
shape of the original bone. However, when col-
lagen is removed the mineral structure left
behind is a weak solid that easily crumbles
into a powder. It must be concluded that the
collagen forms a continuous medium but the
mineral does not. The mineral fills voids in
the collagen.
Collagen has been studied for many years and
much is now understood about its chemistry and
ul trastructure but many unresolved problems re-
main such as the three-dimensional order, the
chemical character of the i ntermol ecul ar cross
links and the location of the cross links (Rama-
chandran [13], Gallop and Paz [14], Veis [15]).
Collagen exhibits many levels of order terminat-
ing in a fibrous structure. The successive
levels of organization can be understood with
the aid of figure 5a from Lees and Davidson [2].
The smallest element is the a-helix which
spontaneously combines in triplets to form a
182
AXIS OF
EXTENDED
CHAIN
\
TRIPLE AXESOF
THREE HELICES
A 2.8 A
3.1 A
(a)
(a) The a-helix is left handed. Each
spot is a residue and repeat length
is 3.1 A.
(b) The a-helix is twisted into a right
handed helix which reduces repeat
length between residues to 2.8 K
and the repeat of the supercoil is
10 times residue length, or 28 K.
(c) Three a-helices are threaded to form
a single tropocol lagen (TC) unit.
(Adapted from Glimcher and Krane, .
1968 [12]).
Fig. 5a. The tropocol lagen unit (from
Lees and Davidson [2] ) .
superhelix molecule, the tropocol 1 agen (TC)
unit. In turn TC forms microfibril ropes, which
in their turn form three-dimensional fibrils.
Fibrils join to form fibers seen in tendon.
1. The basic element, the a-helix chain, has
a molecular weight of about 100,000 and several
types have been identified depending on the
amino acid composition. The aj chain has 1052
amino residues of which 1011 are in triplet sets
where the first term is glycine. The N terminus
has 16 nonhelical residues beginning with an
amino (NH2) group while the C end has 25 non-
helical residues terminating in a COOH group.
Gallop and Paz [14] as well as Hulmes et al . [16]
show the residue map. It is the triplet with
first term glycine that defines the character-
istic pattern through all subsequent hierarchal
levels.
2. Three a-helices are wound into a super-
helix about 300 nm long, 1.5 nm diameter (wet)
and a molecular weight of about 300,000 (Rama-
chandran). This is the basic molecule, the TC
unit. In experiments, the a-helix spontaneously
links up to form collageneous materials which
have important industrial and medical uses but
the product is not the ordered collagen of bone
and tissue. It does demonstrate that the
hydrogen bonded state between the a-helices is
stable and the basis for interaction between TC
units in the higher levels of structure.
3. TC units join in sets that exhibit
"quarter staggering" in tissue as illustrated
in figure 5b, based on a submolecular length
of 67 nm or 234 residues, the D-length (Hodge
[17]). A TC unit is 4.4 D units long and packs
into a structure exhibiting a repeated 0.6 D unit
gap between TC ends. It is a requirement of
Hodge's scheme that no two quarter stagger gaps
can be adjacent, which imposes a severe restric-
tion on the three-dimensional structure that has
not yet been resolved (Segrest and Cunningham
[18]).
4. The nonhelical chains at each end of the
TC unit and some of the adjacent glycine coded
triplets are probably the links between molecule
ends but may also serve to link helices within
a single molecule, i.e., the inter and intra-
molecular links are similar. However, many of
the intramolecular links must be hydrogen bonds,
while much of the intermol ecular links between
parallel molecules must be chains covalently
bonded as side chains to residues in the back-
bone of the molecule.
5. A number of three-dimensional structures
have been proposed for the microfibrils, fibrils
and fibers. We use Smith's five strand rope of
TC units [19] as modified by Miller and Parry
[20]. It is shown in figure 5c as Smith pro-
posed the model with a 67 nm axial stagger, 72°
aximuthal displacement between axes of nearest
neighbors and 43 nm hole between colinear TC
units. It incorporates a repeating pattern
after 5 x 67 nm and a helix with a five-fold
1 — HwA
2— H
3— H
4— 1
0.6 D HOLE REGION
N I C
-A V
N END = 16 RESIDUE LONG NONHELICAL END + TRANSITION HELICAL REGION
C END = 25 RESIDUE LONG NONHELICAL END + TRANSITION HELICAL REGION
D = REPEAT LENGTH OF 234 RESIDUES
(Adapted from Gallop and Paz, 1975 [14])
Fig. 5b. "Quarter" staggering scheme (from Lees and Davidson [2]).
183
40 A
s
CROSS LINK
Fig.
5c. Smith's five stranded rope model
for the microfibril (from Lees
and Davidson [2]).
UNIT CELL 80x 80A
Fig. 5d. Miller and Parry model of the structure
and packing of a fibril (from Lees and
Davidson [2]).
screw axis (five subunits per screw turn char-
acterized by the holes). The pentagonal unit
is 4 nm in diameter when the wet TC unit is
1.5 nm diameter and the lumen is 1.1 nm.
6. Miller and Parry indicated that the five-
strand rope must be twisted to produce a four-
fold symmetrical supercoil, where the holes are
now 90° apart, but there is still a pentagonal
packing of the strands. The colinear TC units are
inclined about 2.5° to the rope axis to accom-
modate the twist.
7. Miller and Parry have deduced a fibril
packing structure based on a square or tetragonal
cell of four microfibrils as in figure 5d. While
Miller and Parry did not say so, the pattern of
successive helical structures suggests that the
microfibrils are coiled about each other while
maintaining the four unit cell.
On the basis of this description, the ultra-
structure of collagen can be schematically rep-
resented as in figure 6 where the microfibrils
Si:
INTERMOLECULAR
CROSS LINK
Fig. 6. Schematic representation of the
ultra-structure of collagen showing
intermolecular cross links.
184
are tied together with cross links of organic
chains. The TC units are probably very stiff
along their major axis. Enemoto and Krimm [21]
calculated a value for the Young's modulus of a
very similar molecule, polyglycine II, which
they found to be 41 GPa, a value far in excess of
any reported value for collagen or even for bone
(26 GPa). Polyglycine II has a triple helix
hydrogen bonded molecule and closely resembles
collagen in its structure (Ramachandran [13]).
The Young's modulus for TC is probably about the
same value, which indicates that the collagen
structure is more yielding than the TC units be-
cause the cross links are softer. Moreover,
the long skinny TC units must bend quite easily
unless they are severely restrained, which makes
collagen tissue softer than its component mole-
cules.
7. Mineralization of Bone
As the tissue mineralizes, HAP crystallites
begin to fill the voids in the quarter stagger
gaps, between microfibrils and between fibrils.
The loci are inferred because they have been
sited definitely only between fibrils (White et
al . [1]). It is our contention that the crystal-
lites form on the cross links until they grow to
a size that fills the voids. It is implied that
the crystallites start with the cross links as
the nucleating seeds, which further implies that
the stereochemistry of the organic chain must be
favorable to the hexagonal habit of HAP. There
is no accepted theory to explain how crystallites
PLATELIKE
CRYSTALLITE
NEEDLELIKE
^CRYSTALLITE
Fig. 7. Schematic representation of
mineralized bone collagen showing
crystallites embedding the cross
1 inks.
form and grow in tissue but some workers indicate
that the process is mediated by a sequence of
chemical stages rather than precipitating from a
saturated solution. Whatever the process it en-
tails a mineralization of the organic cross links
until the crystallite encases the links. In
figure 7 a schematic representation of the
mineralized microfibrils shows the crystallites
embedding the cross links and effectively short-
ening them. There is a cross link from one
microfibril to a crystallite and its continua-
tion on the other side connects the crystallite
to a second microfibril. The connections are
made to TC units in the strands of the micro-
fibrils, hence to side chains of the residues on
the a-hel ix.
The effect of mineralization in bone is to
shorten the cross links and thereby increase the
longitudinal modulus of the matrix. There may
well be additional cross links formed at the
same time or prior to mineralization, but the
cross linking is most likely mediated by a dif-
ferent chemical process than that for mineraliza-
tion. The stiffening of the collagen matrix need
not necessarily be associated with a greater
cross linking density.
8. Ultrasonic Velocity of Mineralized Tissues
In order to apply the theory it has been neces-
sary to estimate the equivalent isotropic sonic
velocity for hard tissues. Since bone and dentin
are anisotropic the estimates are not well based
and the ultimate test and evaluation of this
theory requires a better way to find the iso-
tropic velocity equivalent. Alternatively, it
may be possible to extend the theory to deal with
the anisotropy as Yoon and Katz [22] are attempt-
ing to do but the theory is still incomplete.
Figure 8 shows representative values for the
longitudinal sonic velocity of bone, dentin and
enamel. There are four bone values, one for nor-
mal bovine bone by Lang [23], and three by
Abendschein and Hyatt [24]. The latter include
one normal human bone and two of lower density
from ill people. It is interesting that the nor-
mal human and normal bovine bone values coincide
despite the quite different methods for making
the measurements.
Unlike the mineral filled composites we do not
yet have good values for the elastic properties
of the constituents of bone, particularly bone
collagen. Several Reuss curves were calculated
based on the different published estimates for
the elastic modulus of collagen, usually the
Young's modulus which was converted to the
longitudinal modulus by assuming Poisson's ratio
to be 0.35 (Katz [5]).
Three Reuss curves are shown in figure 8.
The lowest bound is based on the longitudinal
modulus suggested by Currey [3] and Katz [5],
3.75 GPa. The intermediate bound is calculated
from Mason's value [7] measured ul trasonical ly
on kangaroo tail tendon. The upper bound was
obtained by using Reuss 's formalism backward on
the sonic velocity and density of normal bone.
It can be seen from figure 7 that the inter-
mediate and upper curves bound the experimental
data, much like the two bounding curves in fig-
ure 4. The lowest bound is based on a value of
longitudinal modulus based on low strain rate
test data, a technique that characteristically
185
3-
1-
INTERMEDIATE
COLLAGEN
K MODULUS
—[ —
10
—\ —
20
— I —
30
— I —
40
50
— I —
60
70
— 1 —
80
much greater than isolated collagen because in-
timate association with the mineral raises the
collagen's tensile stiffness."
References
[1] White, S. W., Hulmes, D. J. S. , Miller, A.,
and Tinmins, P. A., Collagen-mineral axial
relationship in calcified turkey leg tendon
by x-ray and neutron diffraction. Nature 266,
421-425 (1977).
[2] Lees, S. and Davidson, C. L., The role of
collagen in the elastic properties of cal-
cified tissues, J. Biomech. 10, 475-486
(1977).
[3] Currey, J. D., Three analogies to explain the
mechanical properties of bone, Biorheol ogy
2, 1-10 (1964).
- [4]
[5]
90 100
[6]
VOLUME, PERCENT H YDROXYAPATI TE
Fig. 8.
Ultrasonic velocity in mineralized
tissues (from Lees and Davidson [2]).
yields low values compared to ultrasonic mea-
surement techniques.
The data represented in figure 8 can only be
regarded as indicative and not definitive. In
particular the value of enamel cannot be taken
as properly a member of this class of materials
because it is not a collageneous tissue. Enamel
is formed by a totally different process than
that for bone and dentin. The HAP content of
enamel is so high that the mineral component of
eq. (1) dominates, which will be true for a com-
parable mineralized collageneous tissue. How-
ever, the theory developed here shows that it is
unlikely to have so much mineralization in bony
tissues.
Acknowledgment
S. L. is pleased to acknowledge support from
NSF Grant GH-42515 and National Institute of
Dental Research, NIH, Research Grant Number
ROl-DE-3992. Some of the experimental data was
obtained while at the Department of Materials
Science, University of Amsterdam with funds from
the National Science Foundation and the Univer-
sity of Amsterdam.
Note added in proof:
The authors have recently learned that some
of the concepts presented here were surmised
by McCutchen [25]. In his paper he says "I sug-
gest that collagen is the prime tension carrier
in bone, and that bone's stiffness modulus is
Welch, P. 0., The composite structure of
bone and its response to mechanical stress,
in Recent Advances in Engineering Science,
A. C. Eringer, ed. , Vol. 5, pp. 245-262
(Gordon and Breach,
Katz, J. L., Hard tissue as a composite
material - I. Bounds on the elastic be-
havior, J. Biomech. 4, 455-473 (1971).
Currey, J. D., The relationship between
the stiffness and the mineral content of
bone, J. Biomech. 2, 477-480 (1969).
[7] Mason, P., Viscoelasticity and structure
of keratin and collagen, Kol loid , Z. , A.
Polymere 202, 139-147 (1966T;
[8] Papadakis, E. P., Balanced resonator for
infrasonic measurement of Young's modulus
and damping in flexure, J . Test and Eval .
1, 126-132 (1973).
[9] Lees, S. and Davidson, C. L., Ultrasonic
measurement of some mi neral -f i 1 1 ed
plastics, IEEE Trans. Sonics and Ultra-
sonics SU-24, 222-225 (1977).
[10] Kumins, C. A., Long range effects of
polymer pigment interaction in the solid
state, J. Paint Technology Engineering 37,
Pt. 1, pp. 1314-1336 (1965).
[11] Burhans, A. S., Pitt, C. F., Sellers, R. F.,
and Smith, S. G. , High performance epoxy
resin systems for fiber-reinforced com-
posites. Prelim. Paper, 21st Annual Mtg.
Reinforced Plastics Division, Soc.
Plastics Ind. (1965).
[12] Glimcher, M. J. and Krane, S. M. , in Treati se
on Collagen, B. S. Gould, ed.. Vol. 2
(Academic Press, New York, 1968).
[13] Ramachandran , C. N., in Treatise on Col-
1 agen , C. N. Ramachandran, ed.. Vol. 1,
Chap. 3 (Academic Press, New York, 1967).
[14] Gallop, P. M. and Paz, M. A., Posttran-
lational protein modifications with special
186
attention to collagen and elastin, Physiol .
Rev. 55, 418-487 (1975).
[15] Veis, A., Collagen Biosynthesis, in CRC
Critical Reviews in Biochemistry, Vol. 2,
p. 443 (The Chemical Rubber Co., Cleveland,
Ohio, 1974).
[16] Hulnies, D. J. S., Miller, A., Parry, D. A.
D., Piez, K. A., and Woodhead-Gal loway ,
J. W. , Analysis of the primary structure
of collagen for the origin of molecular
packing, J. MoT . Biol . 79, 137-148 (1973).
[17] Hodge, A. J., in Treatise on Collagen, C. N.
Ramachandran, ed., Vol. 1, Chap. 4 (Academic
Press, New York, 1967).
[18] Segrest, J. P. and Cunningham, L. W. , Unit
figril models derived from the molecular
topography of collagen, Biopoly . 12, 825-834
(1973).
[19] Smith, J. W., Molecular pattern in native
collagen. Nature 219, 157-158 (1968).
[20] Miller, A. and Parry, D. A. D., Structure
and packing of microfibrils in collagen,
J. Mol . Biol . 75, 441-447 (1973).
[21] Enemeto, S. and Krimm, S., Elastic moduli of
helical polypeptide chain structures,
Biophys. J. 2, 317-325 (1962).
[22] Yoon, H. S. and Katz, J. K. , Ultrasonic wave
propagation in human cortical bone, J_. Bio-
mech ■ 9_, I. Theoretical considerations for
hexagonal sysmmetry 407-412; II. Measurements
of elastic properties and microhardness
459-464 (1976).
[23] Lang, S. B., On the anisotropic elastic coef-
ficients of bone and results on fresh and
dried bovine bones. IEEE Trans. Biomed.
Engineering 17, 101-105 (1970).
[24] Abendschein, W. and Hyatt, G. W., Ultra-
sonics and selected physical properties of
bone, Clin. Orthop. 69, 294-301 (1970).
[25] McCutchen, C. W. , Do mineral crystals
stiffen bone by straitjacketing its collagen?
J. Theor. Biol . 51, 51-58 (1975).
187
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed., National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
ULTRASONIC PROPERTIES AND MICROTEXTURE OF HUMAN CORTICAL BONE
Hyo Sub Yoon and J. Lawrence Katz
Center for Biomedical Engineering
Rensselaer Polytechnic Institute
Troy, New York 12181, U.S.A.
The wave propagation in human cortical bone (dried) was investigated, using an
ultrasonic pulse transmission method at room temperature. Firstly, it has been found
that the symmetry of human cortical bone is consistent with the hexagonal system,
based on the ultrasonic velocity measurement and microscopic observations. The five
independent elastic stiffnesses were determined at 5 MHz, and they are (in GPa):
Cii = 23.4 ± 0.31, C33 = 32.5 ± 0.44, c^i, = 8.71 ± 0.13, Cjj = 9.06 ± 0.38,
Ci3 = 9.11 ± 0.55.
Secondly, this study shows that bone filters and polarizes ultrasonic waves.
Thirdly, since, in a piezoelectric medium such as bone, the wave propagation (or
elastic stiffnesses) is modified by the piezoelectric coupling, the piezoelectric
contribution or "stiffening" was calculated for bone, employing the piezoelectric
and dielectric constants of bone reported in the literature. Compared with the
corresponding values of two well-known piezoelectric materials, a-quartz (single
crystal) and "poled" barium titanate ceramic (polycrystal 1 ine ) , it has been found
that the piezoelectric "stiffening" in bone is negligible. Finally, the sound
velocities were measured over the frequency range of 1 to 5 MHz for the transverse
mode and of 2 to 10 MHz for the longitudinal mode. For all the eight independent
modes the ultrasonic velocities are found to increase with increasing frequency,
implying that bone is viscoelastic even at these high frequencies.
Key words: Anisotropy; dispersion; elasticity; human bone; microhardness ; micro-
structure; piezoelectricity; thermodynamics; ultrasound; visco-
elasticity; wave propagation.
1. Introduction
The material symmetry is an important concept
for interpreting and analyzing the structure-
property relationships of a material, from both
theoretical and practical points of view. This
is especially true when one has to deal with many
tensorial properties of a multiphase-composite
material such as bones and teeth. The maximum
use of symmetry elements in the medium simplifies
the analysis enormously and provides a clearer
physical picture for any phenomenon than other-
wise.
It is known that bone has a remarkably well
organized structure, consisting mainly of the
protein (collagen, not well crystallized), the
inorganic phases (crystalline hydroxyapatite and
possibly a type of amorphous calcium phosphate)
and a fluid phase (in vivo). Therefore, one can
expect that several crystal-physical phenomena
occur in bone. It has, in fact, been reported
that bone is both pyroelectric (first-rank
tensor) and piezoelectric (third-rank tensor).
When employing ultrasonic waves of low strain
levels much less than 10"^, as in diagnostic
ultrasound, one must take into account these
interactions between thermal and electrical
properties, and between elastic and electrical
properties, respectively. In other words, it is
necessary to regard bone as a thermodynamic sys-
tem. However, the situation becomes more com-
plicated for a more realistic case than the
above idealized one because bone appears to be
viscoelastic even at ultrasonic frequencies.
It is now clear from the above information
that one cannot treat each physical or thermo-
dynamic property independent of the others, as
has been done in the past, e.g. , [l,2]i. Instead,
one should start with a simple situation when
studying bone, then move to a more complex model
by adding one or more independent variables, as
was followed in the present investigation on
human compact bone. For this purpose it is con-
venient to assign hierarchal levels of struc-
tural organization to bone, that is, molecular,
ul trastructural , microscopic and macroscopic.
The wavelengths of ultrasound employed (3.7 x
10" 1 to 2.2 mm) are of the order of magnitude
appropriate to the microscopic level. This is
the level at which conventional metal lographic
techniques are best suited to study the micro-
structure or microtexture of bone. Note that
the microtexture of human bone is different from
^Figures in brackets indicate literature
references at the end of this paper.
189
that of bovine, canine, or any other mammalian
bone.
The simplest situation for ultrasonic studies
of bone is one in which temperature (room) and
frequency (5 MHz) are kept constant. This cir-
cumstance corresponds very closely to the situa-
tion of plane wave propagation in an elastic
medium and yields a set of elastic stiffnesses
for bone. As the next step, the piezoelectric
contribution was considered for the ultrasonic
wave propagation in bone. So far, bone has been
treated as an elastic dielectric. In order to
determine the existence of a frequency dependence
or dispersion of the ultrasonic velocities at
room temperature, the sound velocities were
measured over the frequency range of 1 to 5 MHz
for the transverse mode and of 2 to 10 MHz for
the longitudinal mode. These results are im-
portant when it is necessary to accurately de-
lineate ultrasonic paths, as in ultrasonic ence-
phalography, monitoring of fracture healing, and
other medical and dental diagnoses since the at-
tenuation of ultrasonic waves in calcified tissue
is substantially larger than that in soft tissue.
2. Theoretical Background:
The Thermodynamics of Bone
In order to derive the relationships between
the mechanical, electrical and thermal proper-
ties of bone, i.e. to establish the thermo-
dynamic aspects of bone, let us, following Voigt
[3], introduce a thermodynamic potential per
unit volume, E, as a function of ten independ-
ent variables: stress T^, electric field
strength E^- and entropy a, combined with the ap-
propriate coefficients, elastic stiffness c^v^
dielectric permittivity e-jj, specific heat C,
piezoelectric stress constant e-jy, thermal
stress coefficient q^ pyroelectric constant p-j
and absolute temperature 0. Thus:
1 E,a r c X 1 S,0 r.r. , I ^
^ c ' S S + ly £ . ■ Ei E-i + y -^-F
(1)
6 qt
+ e^ EiS„ +
IP ' y CE
S,,a +
e p-:
J\
cs
EiO,
where summation over repeated indices is implied,
and i ,j = 1, 2, 3; M, V = 1, 2, . . . , 6. A
superscript indicates the quantity to be kept
constant. Here magnetic effects are not includ-
ed in eq. (1) because they are usually very small
compared to electric effects. Also, bone is
treated as an anisotropic linear elastic solid,
i.e. as a linear elastic dielectric, rather than
as either metallic, ferromagnetic, or ferro-
electric solids. That is, for simplicity the
viscoelastic properties are considered separate-
ly.
Since the propagation of ultrasonic waves
through a medium is an adiabatic process (isen-
tropic), eq. (1) is simplified to
ic^'° S S
2 pv V \
1 y
EiSy
(2)
Differentiation of eq. (2) with respect to Sp
and E-j, respectively, gives the well-known equa-
tions of state (or constitutive equations) for a
piezoelectric medium:
T = c^'^ S
^ yv
eV
i / T,a
S,a
(3)
(4)
which now define stress T^ and electric displace-
ment D.. Note that the coefficients on the lead-
ing diagonal represent the principal effects
while the off-diagonal coefficients (equal to
each other) measure the coupled effects.
For the pulsed ultrasonic measurements of
thickness vibration, the elastic stiffness ap-
propriate for the vibration would properly be
designated as
Dn,Et,a
^y V '
meaning that the normal component of the electric
displacement and the transverse components of the
electric field are constant (usually zero). By
combining Newton's second law of motion with the
constitutive eqs. (3) and (4) and substituting
plane wave solutions uj (= Uj exp i(a)t - k-x))
for the medium of density p, it may be shown that:
3t2
c .
Dn,Et,a
ijk?.
ax .3x,
where :
°n'^t'° ^ Dn'EfO ^ E,a ^
'^i'jk? "-yv
with m being fixed.
e p
my. "ly
,S,a
mm
(5)
(6)
Note that the directions i, n, m are parallel to
one another, and that
Dn'^t'°
D,o
(7)
is, in general, different from c^^^ which is
given by:
D,a E,o ^iy^jv
C = C + r '
yv yv b,CT
= ij
For more details the reader is referred to
references [4-6].
3. Experimental Procedure
A. Specimen preparation
Bone samples were cut with a band saw out of
the right femoral mid-diaphysi s from the cadaver
of a 57-year-old male who died of gastric adeno-
carcinoma. Two wel 1 -oriented specimens in the
shape of a rectangular parallelepiped (0.5461 cm
x 0.8933 cm x 0.6233 cm for X1X2X3 cut and
0.4750 cm x 0.5641 cm x 0.7645 cm for 45° cut),
were then cut using an Isomet low speed saw with
a diamond blade 10.16 cm diameter and 0.03048 cm
thick. Distilled water was used as the lubricant
190
and coolant during cutting. The specimens were
oriented using a straight surface in the medial
quadrant, which is always parallel to the bone
axis (X3 axis) in any femur, taking the axis
along the radial direction and the axis per-
pendicular to both, thus forming a right-handed
rectangular coordinate system. The lateral
quadrant of the mid-di aphysi s was selected for
obtaining the specimens because, for this partic-
ular bone, this quadrant was macroscopically
more uniform than was the rest of the cross
section.
Conventional metal lographic techniques were
then employed for grinding and polishing the
specimens. In order to prepare a pair of paral-
lel surfaces for ultrasonic measurements, the
specimen, mounted on a parallel -face device with
double-stick tape, was ground under a flow of
cooling water, with an AB Handimet Grinder on
which a 600 grit silicon carbide paper strip was
pasted. This was followed by polishing with 0.3
pm alumina slurry on a 20.32 cm diameter bronze
wheel covered with wool broadcloth. Each time
after grinding and polishing, the specimen was
cleaned in an ultrasonic cleaner for 3 minutes.
The orientation of a specimen was checked by
microscopic observation of osteon arrangement in
the transverse cross section.
Each specimen, wrapped with soft tissue papers,
was dried very slowly to avoid undesirable results
such as cracks and distortions. This was done by
keeping the specimen in a glass vial whose plastic
lid was tightly closed, for a day, followed by
placing it first in a desiccator for 10 days, and
subsequently in a vacuum oven at 24 °C and 533 Pa
for 4 days. Finally, the specimen was dried in a
vacuum line equipped with both a mechanical pump
and a diffusion pump at room temperature for 3
days. Before proceeding with the ultrasonic
velocity measurements, the moisture content of
the specimen was stabilized by keeping it in air
until changes in the specimen's dimensions were
recorded as less than 0.1 percent.
B. Micrographs and microhardness
measurements
From the same mid-diaphysi s of human bone,
another cylindrical specimen 0.5 cm long was pre-
pared in the identical manner as described before.
This specimen was used both for obtaining micro-
graphs (75X) of the transverse cross section and
for Vickers (HV) microhardness measurements, em-
ploying a Bausch & Lomb Metal loqraph and a
Kentron Tester, respectively. A typical micro-
structure of the transverse cross section is
shown in figure la. Compare it with that of
young bovine bone shown in figure lb. The hard-
ness of the bone specimen was measured with a
200 g load at locations equally spaced radially
on the four quadrants: anterior, posterior,
lateral and medial. Care was taken to make all
indentations away from Haversian canals or
lacunae.
r
(a) HUMAN
(b) BOVINE
Fig. 1. Microstructures of human and bovine demora, 75X (transverse cross section),
(a) human and (b) bovine.
C. Ultrasonic setup and measurements
A block diagram for the pulse through-trans-
mission technique is shown in figure 2. For each
frequency step, a pair of PZT-5A piezoelectric
transducers (chromium-gold plated on both faces)
was used to transmit and receive the ultrasonic
pulses. For the specimen holder, a pair of
parallel faces of aluminum buffer disc was pre-
pared by hand-grinding it on 400- and 600-grit
silicon carbide paper strips, as explained pre-
viously, and by polishing with a 3 pm diamond
compound; the disc was cleaned ul trasonical ly
after each step. The transducer was cemented on
one face of the buffer disc with salol (melting
point 43 °C), the other face being in contact
with a face of the specimen during the transit-
time measurements. No coupling medium was used
191
RF PULSED OSCILLATOR
(ARENBERC P&- 65UCI
PRECISION ATTENUATOR
lARENBERG ATT-6931
J2^_TRANSM TRANSDUCER
( PZT 5A I
OSCILLOSCOPE
ITEKTRONII 5431
IZ,^ RECEIV TRANSDUCER
(PZT 5A)
WIDE BAND AMPLIFIER
(H P 460 A)
Fig. 2. Block diagram of pulse transmission method.
between the buffer disc and the specimen surface
since dried bone is porous and the specimen volume
was very small. Instead, the acoustic coupling
was formed by lightly pressing down the upper
part of the specimen holder with a screw. For
more details see reference [7].
Pulse transit (or delay) times in the specimen
were measured from the shift of the pulse posi-
tions as observed on the horizontal axis of the
oscilloscope with and without the specimen be-
tween the aluminum alloy buffer discs. The
horizontal (time) axis of the oscilloscope was
calibrated with a Time-Mark Generator 180A. The
accuracy of the pulse transmission method was
checked by measuring the transit times both of
AT cut quartz oscillator plates of known frequen-
cies, and a beryl single crystal whose values
were determined previously by Yoon and Newnham
[8] using a pulse superposition method [9].
4. Experimental Results and Discussion
A. The homogeneity and microtextural
symmetry of human cortical bone
Figure 3 shows that the bone is "intrinsically"
homogeneous within the limit of experimental er-
rors along the radial direction on the four
quadrants. Weaver [10] has also observed a wide
zone of mid-cortical bone of uniform hardness for
autopsy and surgical bone specimens. Kallieris'
results [11] show no significant differences in
hardness between fresh human bones of various
structural designs in different layers of compact
bones. These findings justify somewhat the use of
small specimens for measurements of the physical
or mechanical properties of bone.
Rauber [12] and Dempster and Liddicoat [13]
have shown that bone has different physical and
mechanical properties along and perpendicular to
its long axis. Therefore, the symmetry (or
pseudosymmetry ) of the microstructural texture of
bone may be either orthorhombic, tetragonal, trig-
onal, or hexagonal. Since bone is an opaque,
microstructural composite, neither optical nor
single crystalline x-ray diffraction techniques
can be employed to determine its full textural
symmetry. An ultrasonic wave propagation method
can provide information about both the symmetry
and the elastic (and viscoelastic) properties of
either crystalline or non-crystalline materials.
The microstructure on the transverse cross sec-
tion of bone shows a pseudo-hexagonal close-
BOr
40-
ANTERIOR
Relative scale
80
r 40
• • • • • • •
POSTERIOR
_L
Relative scale
ci. BOi-
■r 40-
LATERAL
Relative scale
80
40
• • •
MEDIAL
Relative scale
Fig. 3. Microhardness (Vickers Hardness) of human femur (dried).
P = periosteum and E = endosteum.
192
packing of Haversian systems or osteons. Pre-
viously, Katz [14] and Katz and Ukraincik [15]
suggested that the arrangements of osteons and
interstitial lamellae could be considered to be
pseudo-hexagonal. Thus, the tetragonal and trig-
onal systems can be eliminated. The ultrasonic
measurements have shown that Cn and C22 or c,^l^
and C55 are equal to each other, thereby leaving
only hexagonal symmetry as appropriate for bone.
B. The elastic stiffness of bone
Table 1 shows the ultrasonic velocities to-
gether with their standard deviations at room
temperature and 5 MHz along various directions
in the human femur specimens. The eight in-
dependent velocity measurements provide four in-
ternal cross-checks. For comparison the cor-
responding sound velocities in a Durango fluo-
rapatite single crystal [16] are included. From
these velocities and the independently measured
mass density p, the elastic stiffnesses of bone
were calculated, as shown in table 2 where the
pseudo-single crystal elastic stiffnesses of
hydroxyapatite (HAP) [15], enamel and dentin
[17] are included, in addition to the elastic
stiffnesses of fluorapatite (FAP) [16]. Note in
the case of the mineralized tissues that the
stiffnesses, in general, increase with the amount
of apatite.
Table 1.
Sound velocities in human compact bone and
fluorapatite crystal at room temperature.
Mode Propagation Displacement Sound velocity (km/s)
direction, direction, Fl uorapati te^ Human femur
N U (dried)
aL
aT
yL
TT3
45L
45Th
45T„
[001]
[001]
[100]
[100]
[100]
[A oi]
[001]
[100] or [010]
[100]
[010]
[001]
1/2 "i]
[0 -1 0]
7.586
3.635
6.842
3.978
3.638
7.020
3.811
4.033
4.18
2.16
3.55
1 .98
2.17
3.86
0.03
0.01
0.02
0.01
0.02
0.03
2.06 ± 0.01
2.24 ± 0.02
See reference [16].
^ 1
1 1 1
aL
1 1 1
1 1
■--.^,^^^451
_
yL
-aT
51 —
45Tv
0
YT3-
<i
1111
45Th
1 1 1
>;
YTh
1 1
0 10 20 30 40 50 60 70
Colatitude, (f) (degree)
80 90
Fig. 4. Angular dependence of the elastic stiff-
nesses of human bone.
while the degenerate transverse mode aT along the
bone axis (X3) is separated out into the two
branches, 45Tv ^ YT3 and ISJ^ ^ yT^^.
In a hexagonal medium there exists, in general,
three pure mode directions (a, g, y) along each of
which one purely longitudinal and two purely
transverse modes of propagation occur, forming a
mutually orthogonal set. Of these the a and y
directions (see table 1) are known from the sym-
metry of the hexagonal medium. The 6 direction,
which lies on a conical surface between the X3
axis and its perpendicular, is given in terms of
the elastic stiffnesses of the material. It has
been found that no 6 direction exists in bone.
It may be that in bone the structural arrangement
of the collagen fibrils and fibers and the osteons
may not allow any direction of high symmetry for
the 6 pure mode. This can explain the observed
phenomena that bone polarizes and filters the
ultrasonic waves.
Table 2. Elastic stiffnesses of calcified tissues and apatite
single crystals at room temperature.
Cuu Fluorapatite Hydroxyapatite Enamel Dentin Bone
(GPa) [16] [15] [17] [17] (dried)
150
5
137
115
37
0
23
4 ± 0.31
C33
185
0
172
125
39
0
32
5 ± 0.44
42
51
39
6
22
8
5
70
8
71 + 0.13
C12
48
8
42
5
42
4
16
6
9
06 ± 0.38
62
2
54
9
30
0
8
7
9
11 ± 0.55
p(g/cm5)
3
2147
3
17
2
9
2
2
1
86
In figure 4 are plotted the elastic stiff-
nesses of bone as a function of angle 4) from the
X3 axis. This shows how the elastic stiffnesses
are interrelated to the orientation in a plane
parallel to the X3 axis. In other words, the
longitudinal modes aL, 45L and yl are related,
C. The piezoelectric "stiffening"
Following Fukada and Yasuda [18] and Fukada
[19], point group 6^ was chosen as the piezo-
electric class of symmetry for bone, which is
also consistent with the pyroelectric measure-
ments on bovine bone by Lang [1]. In table 3 is
summarized the piezoelectric contribution to the
elastic stiffnesses of bone, except for the two
^While it is true that bone does not exhibit
crystalline symmetry in the usual sense, as would
be demonstrated by x-ray diffraction analysis, 6
is the appropriate point symmetry to describe the
microstructural level of the organization of bone
in that it is consistent both with the morpho-
logical observations and with the ultrasonic data
within the precision of the measurements.
193
Table 3. Piezoelectric corrections to elastic
stiffnesses of bone.
Mode
-Dn,Et,a
^Dn.Et.a _ E,a
JJV_
aL
^33
aT
yL
TTh
YT3
cE
45Th
2
633
£33
2
ei5
-11
£11 + £33
1.9 X 10-8
0
0
0
3.9 X 10-7
2.3 X 10-6
mixed modes (45L and 45Ty), employing the piezo-
electric constants for horse femur [19] and the
permittivities of bovine bone by Gundjian and
Chen [20] with the correction for frequency de-
pendence following Liboff and Shamos [21]. Note
that the piezoelectric "stiffening" in bone is
very small and may therefore be neglected within
the experimental errors. Table 4 compares the
values
(cD'° - cE'<')/cE.«
of the typical piezoelectric materials a-quartz
and polarized barium titanate ceramic, with the
corresponding values of bone (see 2. Theoretical
background).
Table 4. Comparison of the piezoelectric contribution
to the elastic stiffnesses of bone, a-quartz and
and BaTi03 ceramic.
E,a\ / E.o
Bone (room temp.) a-Quartz (20 °C) BaTiO, ceramic (25 °C)
Point group, 6 Point group, 32 Point group, 6 mm
(x 10-') (x io-2) (x 10-2)
11
0.18
0.86
0
93
33
0.19
0
17
0
44
82.0
0.072
28
0
12
0.46
-11.0
2
0
13
0.56
0
- 8
4
14
- 0.99
66
0
1.9
0
D. The frequency dependence of
ultrasonic velocities
Figure 5 shows how the eight sound velocities
in dried human bone change with frequency at room
temperature. In all cases the ultrasonic veloci-
ties are found to increase with increasing fre-
quency. All the transverse modes exhibit almost
^ 4.0-
3.5
3.0
2.4
45L
J L
1_L
1 23456789 10
Frequency, MHz
2.2-
2.0-
1.8
45T,
J
Fig. 5.
1 2 3 4 5
Frequency, MHz
Frequency dependence of sound velocities
in human bone (dried).
similar dispersion behavior, while one of the
longitudinal modes (yL) has a steeper slope than
the rest of modes .
Brillouin [22] summarizes the explanations of
the geometric dispersion due to a periodic or
discrete lattice, which were originated by Cauchy
and refined by Powell and Kelvin. This book also
includes Brillouin's own research on the subject.
More recently, Sutherland and Lingle [23] report-
ed the geometric dispersion of acoustic waves by
an elastic-elastic composite of tungsten wires
embedded in an aluminum matrix over the incident
frequency range of 0.63 to 8.57 MHz. They also
194
reported the existence of pass- and forbidden-
bands, in addition to the shift of the incident
frequency upon propagating through the material.
In their case the phase velocity always decreases
with increasing frequency within each pass band.
For a linear viscoelastic solid, e.g. , poly-
methyl methacrylate, Asay, Lamberson and Guenther
[24] reported on the viscoelastic dispersion of
ultrasonic waves, in which the phase velocity
increases with increasing frequency and approaches
a plateau. These results can be explained by re-
laxation phenomenon of "molecular" units (see
Gross [25]).
As mentioned earlier, bone is a hierarchal
composite on many levels of structure. Of prin-
cipal concern here is the microstructural level
of organization. In this case the osteons be-
have as stiff hollow elastic fibers ensheathed
by a compliant viscoelastic material, mainly col-
lagen and protein polysaccharides.
It is clear that much additional data are need-
ed with respect to ultrasonic attenuation in bone
over a wide frequency range as well as spectral
analysis of both transmitted and reflected pulses.
In addition it would be valuable to perform
studies on synthetic composites of collagen and
hydroxyapatite in various combinations in order
to obtain dispersion data from controlled com-
posites. Still from the modelling described
above, it is reasonable at present to explain the
dispersion of the phase velocities in human com-
pact bone in terms of its viscoelastic components.
Acknowledgment
Contribution No. 94 from the Laboratory
for Crystallographic Biophysics; this work
was supported by US PHS through NIDR Grant
No. 5Tl-DE-n7-14.
References
[1] Lang, S. B., Pyroelectric effect in bone
and tendon. Nature 212, 704-705 (1966).
[2] Lang, S. B., Ultrasonic method for measur-
ing elastic coefficients of bone and re-
sults on fresh and dried bovine bones,
IEEE Trans. Bio-Med. Engng. 17, 101-105
TTW: ^ —
[3] Voigt, W., Lehrbuch der Kristal Iphysik
(B. G. Teubner, Leipzig, 1910); reprinted
with an additional appendix, 1928 (Johnson
Reprint Corp., New York, 1966).
[4] Cady, W. G. , Piezoelectricity (Dover Pub-
lications, New York, 1964).
[5] Yoon, H. S. and Katz, J. L., Ultrasonic
wave propagation in human cortical bone -
I. Theoretical considerations for
hexagonal symmetry, J. Biomech. 9^, 407-
412 (1976).
[6] Yoon, H. S. and Katz, J. L. , Ultrasonic
wave propagation in human cortical bone -
III. Piezoelectric contribution, J_.
Biomech. 9, 537-540 (1976).
[7] Yoon, H. S. and Katz, J. L., Ultrasonic
wave propagation in human cortical bone -
II. Measurements of elastic properties
and microhardness , J. Biomech. 9, 459-
464 (1976).
[8] Yoon, H. S. and Newnham, R. E., The elastic
properties of beryl , Acta Cryst. A29, 507-
509 (1973).
[9] McSkimin, H. J., Pulse superposition
method for measuring ultrasonic wave velo-
cities in solids, J. Acoust. Soc. Am. 33,
12-16 (1961).
[10] Weaver, J. K., The microscopic hardness of
bone, J. Bone Joint Surg. 48A, 273-288
(1966).
[11] Kallieris, D., Hartemessungen an frischen
menschlichen Knochen, Z. Rechtsmed. 68,
164-170 (1971).
[12] Rauber, A. A., Elasticitat und Festigkeit
der Knochen (Engelmann, Leipzig, 1876).
[13] Dempster, W. T. and Liddicoat, R. T.,
Compact bone as a non-isotropic material.
Am. J. Anat. 9^, 331-362 (1952).
[14] Katz, J. L. , Anisotropic elastic properties
of calcified tissues, lADR Program and
Abstracts of Papers, p. 71, Abstract No.
73 (1971) (49th General Session, Intern.
Assoc. Dent. Res., Chicago, Illinois).
[15] Katz, J. L. and Ukraincik, K. , On the
anisotropic elastic properties of hydro-
xyapatite, J. Biomech. 4, 221-227 (1971).
[16] Yoon, H. S. and Newnham, R. E., Elastic
properties of f 1 uorapati te , Am. Mineral.
54, 1193-1197 (1969); Yoon, H. S., Elastic
properties of fluorapatite and beryl, Ph.D.
Thesis (Pennsylvania State University,
University Park, PA, 1971).
[17] Lees, S. and Rollins, F. R. , Jr., Anisotropy
in hard dental tissues, J. Biomech. 5^, 557-
566 (1972).
[18] Fukada, E. and Yasuda, I., Piezoelectric ef-
fects in collagen, Jap. J. Appl. Phys. 2»
117-121 (1964).
[19] Fukada, E., Mechanical deformation and
electrical polarization in biological sub-
stances, Biorheology 5_, 199-208 (1968).
[20] Gundjian, A. and Chen, L. L., Standardiza-
tion and Interpretation of the Electro-
mechanical Properties of Bone, IEEE Trans.
Biomed. Engng. 21,, 177-182 (1974T;
[21] Liboff, A. R. and Shamos, M. H. , Solid State
Physics of Bone, in Biological Mineralization,
I. Zipkin, ed.. Chap. 14, pp. 335-395 (Wiley,
New York, 1973).
[22] Brillouin, L. Wave Propagation in Periodic
Structures (Dover Publications, New York,
1953). ^
195
[23] Sutherland, H. J. and Lingle, R. , Geometric
dispersion of acoustic waves by a fibrous
composite, J. Composite Mater. 6_, 490-502
(1972).
[24] Asay, J. R., Lamberson, D. L., and Guenther,
A. H., Pressure and temperature dependence
of the acoustic velocities in polymethyl-
methacrylate, J. Appl. Phys. 40, 1768-1783
(1969).
[25] Gross, B., Mathematical Structure of the
Theories of Viscoelasticity (Hermann &
Cie. , Paris, 1953).
196
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed. , National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
ATTENUATION AND DISPERSION OF ULTRASOUND IN CANCELLOUS BONE
James E. Barger
Bolt Beranek and Newman Inc.
Cambridge, Massachusetts 02138
Measurements of the insertion loss and insertion phase shift for ultrasound
transmitted through sections of cancellous bone from human skull are reported as
functions of frequency over the range extending from 0.3 MHz to 3.0 MHz. The
frequency dependence of insertion loss and of phase speed are both found to be
caused principally by scattering of sound by the blood and fat filled interstices
in the bone matrix. Independent scattering measurements made at all observation
angles confirm the scattering phenomenon. It is concluded that both the high
attenuation and the significant dispersion at frequencies above about 1 MHz will
limit the ability to characterize brain tissue by its backscatter at these high
frequencies. Also, most reported sound speeds in cancellous bone have been calcu-
lated from the time-of-f 1 ight of broad-band pulses, and are therefore group speeds.
These group speeds will exceed the phase speeds by about 15 percent in magnitude.
Keywords: Attenuation; dispersion; skull bone; sound speed; ultrasound.
1. Introduction
The objective of this research was to measure,
as a continuous function of frequency in the range
extending from 250 kHz to 2.5 MHz, both the inser-
tion loss and the phase speed of ultrasound in
cancellous bone. These measurements are especially
Important as design data for ultrasound systems
that are to characterize tissue through skull bone,
which comprises two outer tables of ivory bone and
an inner layer of cancellous bone, called diploe.
This diploe layer is known to contribute most of
the ultrasonic attenuation, as well as pulse wave-
form distortions, at diagnostic system frequencies
Successful tissue characterization involves fre-
quency spectrum analysis, over a rather broad band
of frequency, of the sound scattered by a range-
gated segment of tissue [2]. The attenuation that
occurs when scattered sound is transmitted through
diploe can change the spectrum in the band in which
signal amplitude exceeds noise amplitude by an
amount sufficient to permit spectrum identification
Also, the pulse waveform distortions owing to fre-
quency-dependent sound speed, also called disper-
sion, can substantially reduce the range resolution
of the broadband scattered sound.
The experiments reported in this paper were all
performed on cancellous bone samples obtained by
removing the ivory tables from a 9.5-mm thick sec-
tion of human skull that had been fixed for about
six months in buffered formalin.
The experimental apparatus is described schemati
cally in figure 1. A pulse generator applies a
^Figures in brackets indicate literature
references at the end of this paper.
PULSE
GENERATOR
PDPll /40
RF OUTPUT
TRIGGER
INPUT
Focused
Transducer
Diploe
Section
Glass-Walled
tonk
Ultrosonic Probe at Focus
Fig. 1. Schematic diagram of experimental ap-
paratus for measuring insertion loss and
phase through bone samples.
broadband, high-voltage pulse to a focused trans-
ducer. This transducer generates at its focus a
pulse having a peak pressure of about eight bars
and a bandwidth (defined at 10 dB down points) of
about 300 kHz to 2.5 MHz. The pulse is captured
and digitized at 20 MHz and subsequently transmit
ted to a PDPll/40 computer for spectral analysis.
The ultrasonic waveforms are measured at the
focus of the transducer with a small piezoelectri
197
disc probe 3 mm in diameter and 0.3 rrm thick. The
disc is positioned with its major response axis
colli near with the axis of the ultrasonic beam.
The disc diameter is equal to the diameter of the
sound beam at about its 6-dB down points, so that
the probe output voltage is proportional to the
sound pressure averaged over the sound beam.
The insertion loss and phase of the diploe
sample are functions of the Fourier coefficients
T(f) of the pulse at the focus beyond the diploe
sample and of the Fourier coefficients S(f) of the
pulse also measured at the focus, but without the
diploe sample. The insertion loss in dB and the
insertion phase angle are defined in eq. (1). Both
functions are calculated at 19.5-kHz intervals of
frequency by the computer.
IL = -20 log |T/S| (1)
(})EargT-argS.
The phase speed of sound, Cj, in the diploe
sample can be calculated from the frequency ratio
of the insertion phase angle, ((i/f = (j)'. The inser-
tion phase angle is the increase in phase of sound
transmitted through the diploe layer less the in-
crease in phase of sound transmitted through a water
layer having the same thickness, h, as the diploe
layer. This increase is expressed in eq. (2),
where 41 is the phase angle in degrees and Cq is the
phase speed of sound in water.
<}. = 360 f h (CT' - c-i) . (2)
Solving for the phaee speed in diploe, we have
ci = Cq{1 + <l)' c^/SeOh)-' . (3)
If the phase speed Ci is a function of frequency,
which it will be if ^' is not constant, the trans-
mission is said to be dispersive.
2. Experimental Results
A. Insertion Loss
The insertion loss for a 4-mm thick sample of
diploe as a function of frequency is shown on
figure 2. These data show very little loss at 0.3
MHz, with a sharply increasing loss between 0.7 and
about 1.3 MHz. At higher frequencies, loss in-
-lol 1 1 1 1 1
0 3 07 II 15 19
FREQUENCY (MHz)
Fig. 2. Insertion loss of a 4-rran thick sample of
diploe compared with calculated loss due
to scattering and to absorption (A).
creases more slowly with increasing frequency,
reaching about 40 dB at 2 MHz,
B. Phase Speed
The insertion phase angle for a 4-mm thick
diploe sample is shown as a function of frequency
on figure 3. We see that the phase angle does not
1000
1 1
1 1
Cp - 2800 m/s »y -
800
600
400
^^Cp- 2530 m/s
200
Cp" 2190nv*
1 1
1 1
ol i I ' ^
"0 0 5 10 15 20
FREQUENCY (MHz)
Fig. 3. Insertion phase shift of a 4-mm thick
sample of diploe showing three tangent
regions of different group speed Cg.
increase linearly with frequency. Three different
values of phase angle are noted at three different
frequencies. The phase-frequency ratios are
used together with eq. (3) to calculate the phase
speeds at the three different frequencies. The
results at 0.5 MHz and 1.5 MHz are Ci = 2190 m/s
and Ci = 2530 m/s. Similar data on the 4-rim thick
diploe sample at 3.5 MHz showed c^ = 2800 m/s [3].
This tangent is sketched on figure 3, where it is
seen to match with the data for frequencies above
2 MHz.
C. Scattered Sound
The insertion-loss measurement apparatus was
modified, as follows, to measure the sound scat-
tered by the diploe layer. A 6-mm diameter core
was cut from the skull sample and positioned co-
axial ly at the focus. The probe was then scanned
around a 9-cm radius circle centered on the focus
and including the incident sound axis. The sound
pressure levels at 9 cm relative to the sound
pressure level at the focus are shown on figure 4
as a function of observation angle. The narrow
beam directed back towards the source is sound
reflected from the outer table. The small in-
crease in sound pressure level at 180° is the
transmitted sound. Sound reaching the probe at
all other angles has been scattered by the skull
core.
3. Discussion of Results
A. Insertion Loss
The insertion loss is shown on figure 2 to be a (
rapidly increasing function of frequency. The
scattered sound shown on figure 4 is approximately
omnidirectional. Both of these experimental facts
lead us to suspect that sound is being scattered
by the fat- and blood-filled interstices in the
spongy bone matrix. Omnidirectional scattering is
caused by small discontinuities in the bulk modu-
198
ieo°
Fig. 4. Sound pressure levels measured 9 cm from
a core of diploe caused by 20-ps pulses
of 1-MHz sound incident from the direction
of 0 degrees.
lus of the medium. If the average bulk modulus
of diploe is represented by Bq and the bulk modu-
lus of the material in the interstices by B, we
can write the scattering cross-sectional area a of
a single interstice having volume V [4] as
a = 4^3X-^V2(B-B^/Bq)2 . (4)
We consider a plane wave having intensity I in-
cident upon a layer of diploe that is i units
thick and having lateral area A. The total power,
Wg, scattered out of this slice having volume Aj,
is the sum of power scattered by all N interstices
therein:
N N
Ws = lE CTj = I 4it3x-MaB/B^)2 ^ V2 . (5)
j=0 j
We simplify the expression for scattered power by
observing that the volume of the slice M is equal
to the total volume of scatterers NV, where the
average volume is
N
V = N-l ^ Vj .
j
We define also the normalized variance B of the
interstitial volumes.
6 = V2/V2 (6)
Combining eq. (5) with eq. (5), upon defining the
product AI as the power, W^, incident upon the
slice, we have the following result for the ratio
of power scattered from the slice to power incident
upon the slice.
W^/W. = 4Tr3x-'*3 V (aB/Bjj)25, (7)
Equation 7 is equal to the product ai, where a
is the intensity attenuation coefficient, because
the power balance for the power, W^-, transmitted
through a thin slice is W^ = Wi - Ws.
W^/w. = 1 - W^/W. = e-"^ ^ 1 - al (8)
Dividing by incident power and equating to the
definition of attenuation coefficient, we have the
desired equality.
The average bulk modulus of the diploe sample
was calculated from measured mass density and sound
speed. The result was Bq = pqCo^ = 7.81- IQi" ybar.
The bulk modulus of the interstitial contents was
taken to resemble blood, B = pc2 = 2.54-10io ybar.
The average interstitial volume was calculated
from the average diameter d of the interstices,
namely, 0.6 mm. The probability density distribu-
tion for interstice volumes is not known, but we
will assume it to be a Rayleigh distribution, for
which B = 4/tt.
When these quantities are substituted into eq.
(7), we have for the attenuation due to scattering
ll/l (dB/cm) = 4.34a = 13.2 F** , (9)
where F is frequency in MHz.
The fundamental result, eq. (4), is valid only
for scattering elements that have diameters smaller
than about one-third wavelength. For our sample,
then, valid results occur at frequencies less than
1.3 MHz. At higher frequencies, the scattering
law saturates and changes to a frequency-squared
increase [4]. Therefore, the complete analytical
representation of scattering is given by eq. (10).
Il/l (dB/cm) = 13.2 F** F<F
° (10)
=13.2 F2F2 F>F^ ,
where Fo(MHz) = (ci/3d) 10"^.
In addition to attenuation by scattering, there
is attenuation by absorption. A typical value of
the pressure attenuation coefficient in bone is
1.5 nepers/cm at 1 MHz [5]. The associated inser-
tion loss per cm is ll/i = 13. 2F. The total at-
tenuation, namely, the sum of absorption and scat-
tering attenuation, is plotted on figure 2 for a
4-mm thick section. The agreement is very good
between the experimental and the theoretical
values of loss, indicating general accuracy of the
scattering theory.
B. Sound Power Balance
The sound power balance used to obtain eq. (10)
can be checked with the data shown on figure 4.
The incident sound power, W-j , is given in terms of
the incident sound Intensity, I, and the radius, a,
of the focus, W-j = ira^ I. The reflected sound
power, Wp, is given in terms of the reflected half-
beamwidtn, 9, and the reflected inteREity, Ip, mea-
sured at distance ro, W^- = •iT(rosine)2 l„. Tne
scattered sound power, Wg, is given in terms of
the scattered sound intensity, Ig, measured at
distance rg, = 4TTr§ Ig.
The sound power balance requires power incident,
Wi , to equal the power scattered, Wg, plus power
reflected, Wf, plus power absorbed, Wg. We ob-
served in the preceding section that attenuation by
absorption is equal to attenuation by scattering
at a frequency of 1 MHz. The scatter data on
figure 4 are for this frequency, so we take Wg =
199
Wg. Dividing both sides of the power balance by
the incident power, , we have eq. (11).
Wi
= 8{r^/a)Hl^/l.)
+ (rpSin6/a)2(I^/I.;
(11)
The value of Tq is 9 cm, the value of e is about
5° (at 3-dB down point), and the focal radius is
2 mm. The value of Is/Ii is equal to 10"'*-^ = 5,0-
10"5 and the value of Ip/Ii is equal to 10"^-^ =
1.2-10"^, because the reflected pressure is 19 dB
less than the incident pressure and the average
scattered pressure is 43 dB less than the incident
pressure. Substitution of all parameters into
eq. (11) gives the result (W^+Wf+Wa)/!^! = 0.994.
Since a value of 1.0 confirms the hypothesis, this
result is very good, and the sound power is ac-
curately accounted for by reflection, scattering,
and absorption. The transmitted power was neglect-
ed, since it is scarcely more intense than the
forward-scattered sound.
C. Sound Speed
The phase speed, Cp, is defined in terms of the
wavenumber, k: Cp = lo/k. This is the speed that
an observer of an harmonic wave would travel in
order to see a constant wave pressure. The group
speed, Cg, is defined also in terms of the wave-
number: Cg = dio/dk. This is the speed that an
observer of a "packet" of sound energy, having by
necessity a finite bandwidth, would travel in
order to see a constant wave pressure. The two
speeds are different whenever the medium is dis-
persive or whenever k is not a linear function of
frequency. We see clearly from the data on figure
3 that sound propagation in diploe is dispersive.
This result is consistent with our finding that
attenuation is due mainly to scattering, for sound
propagation is then dispersive [6].
Experimental measurements of sound speed in
bone, including diploe, have almost universally
used the time-of-flight of a sound impulse from
which to calculate sound speed. This method
yields the group speed, although experimenters
have not generally noted this. Their results can-
not properly be compared with measurements of
phase speed in diploe, unless the difference is
accounted for.
The definition of group speed yields eq. (12),
which expresses the group speed in terms of phase
speed.
Cp df
1- 1
(12)
The phase speed calculation from our data on the
4~mm skull sample (fig. 3 and ref. [3]) are plotted
on figure 5. We see that a simple frequency-power
law fits the experimental data quite well, accord-
ing to Cp f", where n = 0.131. In this case
eq. (12) gives
n)-i = 1.15 c^
This calculated group speed is plotted on figure 5
together with the experimental values of phase
speed.
MEASURED GROUP SPEED,,
CALCULATED GROUP SPEED Cg
MEASURED PHASE SPEED Cp
FREQUENCY (MHz)
Fig. 5.
Measured values of phase speed and group
speed in diploe shown with group speed
calculated from the phase speed.
The sound speed of the 4-mm diploe sample was
measured by the time-of-flight method, using three
pulses containing most of their energy between 0.3
to 0.4 MHz, 1.0 to 1.5 MHz, and 2.0 to 3.0 MHz
[3]. These measurements give the group speed, and
the values are plotted also in figure 5. It can
be seen that the measured values of group speed
are higher than measured values of phase speed,
and that they agree rather well with the calculat-
ed values.
D. Use of Focused Beam
A focused beam was used to make the measurements
reported herein because only in this way can a nar-
row beam be generated several transducer-diameters
away from the transducer. The scattering of sound
by the bone adversely affects the beam shape at the
focus only if a parameter g = /tT k^aL exceeds
unity [7]. The mean square deviation of the index
of refraction is v^, the average diameter of a
scatterer is a, and the distance traversed by the
sound beam is L. For the 4 mm thick diploe sample,
the value of the parameter g is only about 0.01,
so that no sensible beam distortion or phase aber-
ration is caused at the focus by the scattering in
the diploe layer.
4. Conclusions
We conclude that scattering of diagnostic ultra-
sound by the blood- and fat-filled interstices in
diploe dominates sound attenuation at frequencies
above about 0.7 MHz and also introduces dispersion.
The detrimental effects of large diploe attenua-
tion can be largely avoided by using interrogation
pulses that contain no frequency components greater
than about 0.7 MHz. Attenuation due to scattering
in the diploe studied equals the attenuation due
to absorption at 1 MHz and exceeds it at higher
frequences.
The detrimental effects due to diploe disper-
sion are confined only to the higher frequency
range, where scattering dominates attenuation.
Therefore, if low-frequency pulses are used, there
will be little dispersion. If, however, high
frequencies are necessary to obtain trans-skull
tissue characteristics by spectral analysis, dis-
persion can significantly reduce range resolution.
For example, if a pulse contained energy in the
band from 0.5 to 2.0 MHz, and was transmitted
through a 6-mm thick diploe layer, the depth
200
smearing in the brain due to dispersion would be
0.73 mm, but the nominal depth resolution of the
same 1.5-MHz bandwidth pulse in an undispersive
water- like medium is 0.5 mm.
We find that almost all reported sound speeds
in bone have been obtained by the time-of-flight
pulse method and are, therefore, actually group
speed. There is no evidence that sound propaga-
tion in ivory bone is dispersive, so these data
are equivalent to phase speed. Since we find
sound propagation in diploe to be dispersive, most
published sound speed values in diploe will be
higher than correct values for phase speed.
References
[1] White, D. N., Clark, J. M. , Curry, G. R. , and
Stevenson, R. J., The Effects of the Skull
Upon the Spatial and Temporal Distribution of
a Generated and Reflected Ultrasound Beam
(Ultramedison, Kingston, Ontario, 1976).
[2] Lizzi, F. L. and Laviola, M. A., Ultrasonic
Spectral Investigations for Tissue Charac-
terization in Ultrasound in Medicine, Denis
White and Ralph Barnes, eds.. Vol. 11, pp.
427-39 (Plenum Press, New York, 1976).
[3] Fry, F. J. and Barger, J. E., Acoustical
properties of human skull, to be published
in J. Acoust. Soc. Amer.
[4] Mason, W. P. and McSkimin, H. J., Attenua-
tion and scattering of high frequency sound
waves in metals and glasses, J. Acoust. Soc.
Amer. 19, 472 (1947).
[5] Goldman, D. E. and Hueter, T. F. , Tabular
data of the velocity and absorption of high
frequency sound in mammalian tissue, J. Acoust.
Soc. Amer. 28, 35-37 (1956).
[6] Howe, M. S., Wave propagation in random media,
J. Fluid Mech. 45, 769-83 (1971).
[7] Chernov, L. A., Wave Propagation in a Random
Medium (Dover Publications, Inc., New York,
1967).
201
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed., National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
TRANSKULL TRANSMISSION OF AXISYMMETRIC FOCUSED ULTRASONIC BEAfIS
IN THE 0.5 TO 1 MHZ FREQUENCY RANGE:
IMPLICATIONS FOR BRAIN TISSUE VISUALIZATION, INTERROGATION, AND THERAPY
F. J. Fry
Ultrasound Research Laboratories of the
Indianapolis Center for Advanced Research
Indianapolis, Indiana 46202, U.S.A.
In order for ultrasound to become a practical clinical technique for diagnosis of
many intracerebral diseases in the adult human, it has been necessary to answer ques-
tions concerning insertion loss of skull, temporal and spatial characteristics of
beams operating in the 0.5 to 1.0 MHz range (the requisite frequency is a function
of the particular adult skull) have a maximum single pass skull insertion loss of
nearly 10 dB (20 dB for a pulse echo system) which can be handled with present tech-
niques to provide adequate signal strength from normal and pathological features of
brain. In this frequency range, for skulls in our studies, the appropriately se-
lected frequency is unchanged after double pass skull transmission, the 6 dB beam
width is increased by a maximum of 40 percent and the beam focus is shifted laterally
by a maximum of 3 mm. Resolution of string targets or live brain targets has been
demonstrated to be in the 2 to 3 mm range at 1 MHz and at 0.5 MHz it appears that
4 to 6 mm resolution can be achieved.
A high intensity focused ultrasonic beam (1 MHz) has been transmitted through an
excised adult skull and used to produce a focal thermal flaw in lucite. This simu-
lation test indicates that the induction of transkull focal lesions in live adult
brain may now be possible.
Keywords: Axisymmetric; beams; focal thermal flaw; skull transmission; ultrasound.
1. Introduction
In the adult human skull the diploe layer is a
dominant factor in determining the magnitude of
insertion loss for a transmitted ultrasonic beam.
Insertion loss characteristics for human skull as
a function of frequency have been documented in
the literature [1-4]^ but only recently has the
acoustic character of the specific skull layers
been more clearly defined [5]. These measure-
ments indicate that frequencies less than 0.5
j MHz are necessary to have a single pass insertion
I loss less than 10 dB for some adult human skulls,
i Skulls we have studied with a loss of this magni-
i| tude at a given frequency will transmit an ultra-
I sonic signal of this same frequency with minimal
jj temporal distortion. For infants, the skull in-
f sertion loss at ordinarily used diagnostic fre-
quencies (2.25 MHz) is less than 20 dB since
there is essentially no diploe layer. Since the
two-way insertion loss is twice the one-way loss
in a pulse echo regime, the 20 dB maximum skull
insertion loss (at the appropriate frequency) is
not a serious impediment to pulse echo visual i-
^Figures in brackets indicate literature
references at the end of this paper.
zation and interrogation of brain features.
Preservation of temporal coherence in the wave
is also an important feature and this can be
achieved when the appropriate ultrasonic fre-
quency range is selected.
Use of an axisymmetric focused beam in pre-
vious studies indicated that preservation of the
beam spatial focal character was achieved in a
qualitative sense and that the maximum lateral
displacement of the beam was of the order of
3 mm. Studies with phased array systems indi-
cated a need for phase compensation at the re-
ceiver array to correct for local skull thick-
ness variations which change the arrival time of
the waves at the receiver elements [6-8]. Such
variations also lead to variable insertion
1 osses over the skul 1 .
This report covers studies in which low in-
tensity axisymmetric focused ultrasonic beams
suitable for diagnostic medicine have been
transmitted through excised adult human skull
sections. The beams have been studied for their
temporal and spatial preservation as well as
displacement after skull transmission.
Transkull B-scan presentations of physical
targets and live human brain indicate consid-
erable potential for clinical diagnostic
medicine.
203
No previous literature seems to exist which
indicates the possibility of transmitting an
intense focused ultrasound beam through an adult
human skull which would be capable of producing
focal lesions in brain. A series of experiments
reported here shows that a high intensity fo-
cused ultrasonic beam of a frequency appro-
priate for a given skull can be transmitted
through the excised adult human skull and pro-
duce a thermal focal "lesion" in lucite [9].
Previous studies show that lucite, in the
physical model sense used here, can simulate
1 ive brain tissue [10] .
2. Materials and Methods
Measurement of frequency dependent attenua-
tion loss in skull was made with a system de-
scribed elsewhere [5]. A schematic diagram of
the essential components is shown in figure 1.
Fig. 1. Schematic diagram for skull insertion loss
measurements as a function of frequency.
PA =
pul ser
T =
focused transducer
S =
skull (formalin fixed)
P =
probe
D =
digitizer (8 bit 100 MHz Biomation
8100)
C =
computer (PDP-11/45)
PP =
printer/ploter (Versatec 1100)
x-y-z =
(3 orthogonal) motion coordinate
system
Sound from the transducer passes through the
skull section under study and is received on a
3 mm diameter ferroelectric probe (10 MHz fun-
damental resonance). The acoustic signal from
the probe is analyzed by an FFT (Fast Fourier
Transform) method in the skull and non-skull
case, and the difference of the Fourier coeffi-
cients at each frequency is taken as the inser-
tion loss at that frequency.
Measurements of beam configuration and dis-
placement after skull transmission were made
by moving the probe in a direction perpendicu-
lar to the sound transmission axis. The skull
is also mounted on a coordinate system so that
it can be moved in 3 orthogonal directions.
Rotational axes are also provided so that the
central ray of the transducer can be angulated
with respect to the skull surface.
B-scans of physical models or patients were
made with the apparatus shown schematically in
figure 2. This apparatus provides a mechani-
Fig. 2. Schematic diagram for B-scan imaging.
PA = patient
T = tranceiver
W = water bath
X = mechanically driven linear motion
P = pul ser
A = ampl if ier
D = detector
SC = scan converter
M = monitor
cally driven linear scan path. Such a path is
very limiting for scans involving skull since
it is important to approximately follow the
skull contour so that the central transducer
ray is normal to the skull surface for minimal
insertion loss. In order to have minimal beam
displacement, this normal angulation configu-
ration should be preserved.
For the high intensity focused beam study
involving production of thermal "lesions" in
lucite (lucite used here as a brain simulation
target) the system configuration is shown in
figure 3. A 1 MHz high power transducer [11]
with a 6 dB down (ultrasound intensity) beam
diameter of 2.2 mm was used for this part of
the study. The skulls used were obtained from
Fig. 3.
Schematic diagram for high power
transkull test.
T = high power focused transducer
S = skull (formalin fixed)
L = lucite block (2 inch x 2 inch x
2 inch)
W = temperature-controlled water bath
PA = power amplifier (1 kW)
TG = timing gate
FS = frequency source
204
patients at autopsy and were immediately placed
in Ringer's solution and stored in the refrig-
erator. By the time the experiments reported
here were conducted, the skulls had been placed
in formalin and stored at room temperature.
The storage procedure has been shown to provide
the necessary conditions for long-term stabil-
ity of the acoustical properties of skull
starting from the fresh excised state proper-
ties [12].
3. Results
One particular adult human skull has been
selected from a group used in the preparation of
an extensive manuscript [5] on acoustic properties
of skull to illustrate the frequency filtering
properties of adult skull having a diploe layer.
Figure 4 shows the free field probe received
waveform transmitted from a focused transducer
with a damped resonant frequency of 2.5 MHz.
When the skull is inserted between the trans-
ducer and the probe the signal strength is
greatly attenuated, as will be described later.
A
•| 0. 5 ys/division
Fig. 4. Probe received waveforms (using setup
shown on fig. 1 with an oscilloscope
connected in fromt of D).
B = without skull (2.5 MHz)
A = with intervening skull (skull filter-
ing permits 0.56 MHz transmission)
and the dominant transmitted frequency is 0.56
MHz. When a 0.5 MHz frequency damped pulse is
transmitted through the skull, reflected from
a glass plate and passed again through the
skull, the received waveform recorded from the
original transducer is shown on figure 5. The
double pass transmission loss is 10 dB for this
skull. Use of ultrasonic frequencies in this
range (0.5 MHz) is an advantage because of the
temporal coherence preservation.
Spatial aspects of beam displacement and
distortion after skull transmission for a 0.5
MHz focused beam is demonstrated on figure 6.
These beam aspects have been studied with skull-
to-probe distances from 2 to 10 cm, which is the
range needed for brain target delineation in the
intact skull case with the live patient. In
these two adult skulls the maximum beam dis-
A
i 0.5 ys/division
Fig. 5. Transceiver waveforms after reflection
from glass plate (using setup shown on
fig. 1 with an oscilloscope connected
in front of D) .
A = without skull (0.5 MHz, distortion
is due to the electrical driving
network)
B = after skull transmission (0.5 MHz,
skull has filtered out the high
frequency components in the A
waveform but the 0.5 MHz fundamental
has been preserved.
205
■100
■ 90
■80
•70
1-60
50
1-40
30
500 kHz tranducer
4 mm diameter disc probe
free field plot
L = 2 cm (skull in place)
L = 4 cm
L = 6 cm
L = 8 cm
F = L = 10 cm
Fig. 6. Beam distortion and displacement after single pass skull transmission as a function of probe-
to-skull distance (L) for a fixed transducer-to-probe distance (20 cm). Lateral beam plots
and beam displacement obtained by moving the probe for the field plots (using setup shown on
fig. 1 with oscilloscope connected in front of D to display the receiving probe output as a
function of space coordinate). Insertion loss variations for skulls 6 and 8 were within
± 0.1 dB of the mean insertion loss value for all skull-to-probe distances L. Curves for
each L distance are shown vertically displaced on the graph for reading convenience.
placement from the free field condition is 2.5
mm for any probe-to-skull position, and the
maximum insertion loss variations for the same
range were within ± 0.1 dB of the mean value for
all axial distances studied. These studies also
show that the 6 dB beam width is increased by a
maximum of 40 percent. Other studies show [5]
that when a focused transceiver is used with the
probe as a target, the 6 dB beam widths de-
crease in all cases over the 6 dB beam widths
in the single pass transmission case with the
probe as receiver.
On the basis of temporal and spatial coher-
ence and beam displacement, a 0.5 MHz frequency
is appropriate for adult human skulls. The in-
sertion loss characteristic as a function of
frequency for adult human skull, shown on figure
7, indicates the desirability of this frequency
(0.5 MHz) from a system sensitivity viewpoint if
skull insertion loss is to be minimized. Given
these indicators for using 0.5 MHz ultrasonic
frequency for transkull visualization and inter-
rogation of brain, the important remaining aspect
is resolution. Within the scope of these studies,
using a very preliminary transceiver prototype
(0.5 MHz) and no signal processing to provide
image enhancement, the scan of a string target
without intervening skull is shown on figure 8.
Although the image quality is not ideal, it can
be seen that targets spaced 5 mm apart in real
space can be resolved. Note also that range
resolution has not been optimized since this
prototype transceiver is not highly damped.
+->
+->
Fig. 7.
28-
24
20
16
12
0.6 1.0
Frequency, MHz
1.4
1 .8
Insertion loss as a function of frequency
for single pass skull transmission (skull
6 section - formalin fixed).
206
Skull
Fig. 8. Monofilament nylon string target (0.01
inch string diameter) with 0.5 MHz
focused transducer visualized with
system shown on fig. 2.
A = without skull lowest set of 3 x 5
array has strings 1 cm apart on
centers .
B = with skull
This resolution will be improved with a larger
aperture angle, and with appropriate signal
processing this resolution will be apparent
over a wide dynamic range of signal strengths.
When the skull is interposed there are targets
which show evidence of lateral resolution simi-
lar to that in the non-skull case. Note that
because of this scanning system's inability to
follow the skull contour not all strings can be
visualized. A scan mode in which the scanning
transducer axis is maintained perpendicular to
the skull surface at the point of passage of
the central ray of the focused beam will resolve
this problem.
Our studies include a transkull visualization
ofa glioblastoma in a live adult patient in
which an x-ray transaxial tomographic scan was
also made. Results of these scans are shown on
figure 9. In this case the skull insertion loss
was small enough so that a 1 MHz focused trans-
ceiver could be used (this unit is somewhat more
optimal than the 0.5 MHz prototype). The tumor
tissue was verified by pathological examination
at the time of surgery. The computerized axial
lateral
ventricles
presumed
tumor
region
EMI scan, live patient
focus
marker
lateral
ventricl e
hard
gristle
surgical ly
excised
patholog-
ically
verified
tumor
tissue (1 )
Fig. 9. B-scan of live human patient brain tumor
compared to CAT (computerized axial
tomography) scan. These scans are of a
horizontal section through the brain.
The ultrasound scans were made with the
system shown on fig. 2.
A = CAT scan
B = ultrasonic scan
tomographic scan showed presumed evidence of a
tumor on normal brain structural grounds (ven-
tricular outlines and displacement), but not on
a tissue density difference basis. The ultra-
sound scan clearly shows tissue differentiation.
The linear motion scan mode used does not follow
the skull contour so that the patient-transceiver
angle was selected to permit best viewing over
the tumor region. Spatial resolution of some
targets in the lateral direction is 2 mm and in
the longitudinal direction is 3 mm. It is an-
ticipated that resolution using an appropriate
0.5 MHz transceiver will be no worse than twice
these dimensions.
The above material is relevant to ultrasonic
regimes for transkull diagnostic purposes and
as such uses ultrasonic intensities appropriate
to that usage (low milliwatt average intensity
range). The subsequent material covers experi-
ments demonstrating transmission through excised
adult human skull of intense ultrasonic focal
beams suitable for such interactions as focal
lesion production in brain.
For this study, the high power transducer was
used to produce focal thermal "lesions" in the
lucite block brain tissue simulator [10]. These
lesions were first produced with no skull inter-
vening in the acoustic path. The human skull
section was then inserted in the acoustic path
207
and focal lesions were produced of a size simi-
lar to those produced in the non-skull case.
A number of focal thermal lesions were produced
at various distances from the lucite surface
(typically 2 cm deep). The skull intervening
beam plot compared to the free field case is
shown on figure 10. The single pass insertion
loss was 12 dB for this skull at the 1 MHz
operating frequency of the transducer. More
extensive coverage of this work is provided in
other material [9] .
2 10 1
Distance, mm
Fig. 10. Lateral beam plot obtained with high
intensity focused ultrasonic trans-
ducer, with and without intervening
skull. Measurements made with systems
shown on fig. 1 .
A = without skul 1
B = with skull, the beam axis was
shifted 2 mm laterally from the
free field position (A) and the
insertion loss was 12 dB so no
direct intensity comparisons
should be made from this figure.
4. Conclusion
This work has shown that for ultrasound to
have a significant impact in medicine for diag-
nosis and/or therapeutics of brain disorders, it
is necessary to use ultrasonic frequencies in the
0.5 to 1.0 MHz range. In this range, the single
pass insertion losses can be held to a maximum
of 10 to 12 dB and the temporal coherence of the
wave can be maintained. It has also been shown
that the focused beam configuration can be main-
tained after skull transmission so that the ad-
vantages of resolution and intensity gain can be
utilized. In diagnosis, these advantages lead
to excellent pictorial images of brain struc-
tures and ability to perform tissue interrogation
on temporarily undisturbed waves due to skull
transmission. For potential therapeutics in
brain, these ultrasonic techniques offer the pos-
sibility of transkull ablation of tumor tissue
or other tissues such as that involved for focal
epilepsy and modification of blood-brain barrier
for enhancing chemotherapeutics of tumor therapy
or other brain disorders.
When appropriate ultrasonic frequencies and
focal beams are used, it seems apparent that a
significant impact can now be made in brain diag-
nosis and possibly therapeutics with ultrasonic
devices and techniques.
Acknowledgments
Work reported here was supported by NIH Grant
No. NOl-NS-3-2319, NSF Grant No. APR75-14487 and
The Indianapolis Center for Advanced Research.
References
[I] Heuter, T. F., Cavlieri, A., Langmuir, B.,
Butkus, W., Kyrazia, D., Ballantine, H. T. and
Bolt, R. H., The Detection of Intracranial
Tumors by Use of Ultrasound, Quarterly Progress
Report, Acoustics Laboratory, Massachusetts
Institute of Technology, Cambridge, Mass.,
pp. 38-46, July/September (1951).
[2] Goldman, D. E. and Hueter, T. F., Tublar data
of the velocity and absorption of high fre-
quency sound in mammalian tissue, J. Acoust.
Soc. Amer. 28 (1), 35-37 (1956).
[3] Martin, B. and McElhaney, J. F., The Acoustic
Properties of Human Skull Bones, in Biomedical
Material Research, Vol. 5, pp. 325-333 (John~
Wiley & Sons, Inc., New York, 1971).
[4] Smith, J. W., Phillips, D. J., von Ramm, 0. T.
and Thurstone, F. L., Real Time B-Mode Echo-
Encephalography, in Ultrasound in Medicine,
Vol. 2, D. White, ed., pp. 373-381 (Plenum
Press, New York, 1976).
[5] Fry, F. J. and Barger, J. E., Acoustical
Properties of Human Skull (submitted to
J. Acoust. Soc. Amer. , 1977).
[6] Phillips, D. J., Smith, S. W. , von Ramm, 0. T.
and Thurstone, F. L., A Phase Compensation
Technique for B-Mode Echo-encephalography , in
Ultrasound in Medicine, D. White, ed.. Vol. 1,
pp. 345-404 (Plenum Press, New York, 1975).
[7] Smith, S. W., Mitter, E. B., von Ramm, 0. T.,
and Thurstone, F. L., Signal Processing Tech-
niques for Improving B-Mode Echo-encephalog-
raphy, in Ultrasound in Medicine, D. White, ed.
Vol. 1, pp. 405-414 (Plenum Press, New York,
1975).
[8] Phillips, D. J., Smith, W. W. , von Ramm, 0. T.,
and Thurstone, F. L., Sampled Aperture Tech-
niques Applied to B-Mode Echo-encephalography,
in Acoustical Holography, N. Booth, ed.. Vol. 6,
pp. 103-120 (Plenum Press, New York, 1975).
[9] Fry, F. J., Transkull Transmission of an In-
tense Focused Ultrasonic Beam, Ultrasound Med.
Biol ■ 3 (2,3), 179-184 (1977).
[10] Lele, P. P., Irradiation of plastics with
focused ultrasound: a simple method for eval-
uation of dosage factors in neurological appli-
cations, J. Acoust. Soc. Amer. , 34 (4), 412-
420 (1962y~ ~
[II] Fry, F. J., Heimburger, R. F., Gibbons, L. V.,
and Eggleton, R. C, Ultrasound for Visualiza-
tion and Modification of Brain Tissue, I . E . E. E.
Trans. Sonics and Ultrasonics SU-17 (3), 165-
169 (1970y: ~ ~~
[12] White, D. N., Clark, J. M., Curry, G. R., and
Stevenson, R. J., The Effects of the Skull Upon
the Spatial and Temporal Distribution of a
Generated and Reflected Ultrasonic Beam, avail-
able from Ul tramedi son. Box 763, Kingston,
Ontario, Canada (1976).
208
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed. , National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
SOME ADVANCES IN ACOUSTIC IMAGING THROUGH SKULL
S. W. Smith
Food and Drug Administration
Rockville, Maryland 20852, U.S.A.
D. J. Phillips
Center for Biomedical Engineering
University of Washington
Seattle, Washington 98105, U.S.A.
0. T- von Ramm and F. L. Thurstone
Department of Biomedical Engineering
Duke University
Durham, North Carolina 27706, U.S.A.
Previous attempts to image the adult brain through the skull using diagnostic
ultrasound have resulted in images of poor lateral resolution and limited dynamic
range.. The skull can be modeled as an acoustic lens whose attenuation increases
rapidly above 1 MHz and whose thickness variations introduce phase aberrations on
the order of several wavelengths across the transducer aperture. Statistical analy-
sis of skull thickness data indicate that an electronic sector scanner using a 1 MHz
linear array transducer is less sensitive to these effects of the skull than tradi-
tional pulse echo systems operating at higher frequencies. Representative ultra-
sound tomograms of the brain are shown. In addition, water tank experiments are de-
scribed in which the skull phase aberration, ^{x), was measured on line and removed
by incorporating a compensating phase variation, -<t>(x), into the transmit and re-
ceive timing of a digitally controlled real time phased array imaging system. Pre-
liminary results show that the lateral resolution of the imaging system is restored
in both the transmit and receive modes.
Key words: Brain scanning; echoencephalography ; neurology; phase compensation.
1, Introduction
Since the initial development of B-mode ultra-
sonography, there have been many attempts (Fry et_
al . [1]^; Brinker and Taveras [2]) to produce
TiTgh quality cross-sectional images of the brain
through the intact adult skull. Real time B-mode
echoencephalography was initially demonstrated by
Somer in 1968 using phased array techniques. De-
spite these efforts, B-mode echoencephalography
is seldom used clinically because of the poor
quality of the images obtained with current tech-
niques. One reason for the immediate widespread
application of computerized tomography in the
head has been the failure of B-mode ultrasonic
imaging of the brain. The intervening presence
, of the skull results in poor signal to noise
ratio, limited dynamic range and reduced lateral
resolution.
^Figures in brackets indicate literature
references at the end of this paper.
Three characteristics of skull bone are re-
sponsible for poor quality brain images: (1)
substantial attenuation of diagnostic ultrasound
by the skull bone; (2) rapid increase of that at-
tenuation with the frequency of ultrasound; and
(3) the variation of thickness of the inner table
of the skull. In order to optimize B-mode ultra-
sound images of the brain, these aberrating ef-
fects of the skull bone must be overcome. By a
proper choice of transducer frequency, aperture
size, image format, and the use of effective
signal processing techniques, significant im-
provement may be achieved in ultrasound tomograms
of the brain. In this manuscript these three
problems of acoustic imaging through the skull
will be examined. In addition one promising
technique will be described in which the effects
of phase variations due to the skull have been
removed by a phase compensation process imple-
mented with the real time, swept focus, phased
array imaging system under current evaluation at
Duke University (von Ramm and Thurstone [3]).
209
2. Ultrasound Attenuation of the Skull
A propagating ultrasonic pulse is strongly at-
tenuated after two passages through the skull by
such phenomena as diffractive scattering, mode
conversion, absorption, and the impedance mis-
match at the skull-brain interface (White [4]).
Previous measurements (Hueter [5]; Smith et al .
[11]; Fry et al . [6]) indicate that the attenua-
tion of ultrasound by the skull is strongly fre-
quency dependent increasing from approximately
10 to 20 dB/cm at 1 MHz to 50 to 60 dB/cm at
2 MHz.
In figure 1 the solid curve shows skull at-
tenuation data from Hueter [5] normalized to a
1 cm thickness. The dashed line shows similar
data which we obtained from a preserved segment
of the adult skull. The data indicate that a
2 MHz signal is attenuated approximately 50 dB/cm
of travel through skull bone. The attenuation at
1 MHz is approximately 12 dB/cm. Much of the
energy lost from the interrogating pulse returns
to the imaging system as unwanted acoustic noise.
FREQUENCY (MHz)
Fig. 1. Attenuation of skull bone versus
frequency.
Such a high background reduces the signal to
noise ratio, obscuring low level echo information
and restricting the useful dynamic range of the
imaging system. The result is usually a high
contrast B-mode image of the brain in which only
the strong specular echoes are displayed. From
the figure it can be seen that using a 1 MHz
transducer substantially reduces the problem of
high attenuation due to the skull.
The rapid increase of attenuation with fre-
quency also degrades the lateral resolution of a
B-mode imaging system (Smith et al . [11]). For
a typical broadband diagnostic transducer, the
ultrasonic pulse contains significant amounts of
energy above and below the center-frequency.
After two passes through an adult skull the high
frequency content of the pulse has been signifi-
cantly reduced relative to the low frequency
components. The center frequency of the ultra-
sound is effectively shifted to a lower fre-
quency. For a fixed transducer aperture, such a
frequency shift will result in degraded lateral
resolution. Our experience with linear array
transducers and broadband piston transducers
has indicated that an interrogating 2 MHz pulse
will show a center frequency of approximately
.8 MHz after two passes through the temporal
region of an adult skull. This results in more
than a twofold loss in resolution capability in
the lateral dimensions. A 1 MHz ultrasonic pulse
is also shifted to approximately .8 MHz but the
resultant loss of resolution is not as signifi-
cant since the original transducer aperture will
normally be larger at 1 MHz. For acoustic imag-
ing through the skull, a 1 MHz transducer would
appear to be a reasonable compromise considering
the requirements of system resolution, adequate
sensitivity and transducer size.
3. Thickness Variation of the Skull Bone
Of equal importance is a consideration of the
effects of the thickness variation of the skull
bone on image quality. In the temporal and
parietal areas of the skull where most neuro-
logical ultrasound examinations are made, the
inner table of the skull bone undergoes variation
in thickness on the order of 1 to 2 mm. It has
been shown that phase variations are introduced
across the transducer aperture when acoustic
energy propagates through such a section of skull
(Phillips et al. [8]). Figure 2 (Phillips et al .
[8]) shows two elements of a one dimensional
transducer array simultaneously transmitting an
acoustic pulse through a section of skull. (An
analogous situation would apply in the receive
mode.) Since the acoustic velocity differs be-
tween bone and soft tissue, variations in skull
thickness will produce relative changes in the
phases of the acoustic wavelets emerging from the
inner table of the skull. The velocity of sound
has been reported to be approximately 3050 m/s
in skull bone (Martin and McElhaney [9]) while
the velocity in brain has been measured to be ap-
proximately 1540 m/s (Wells [10]). The phase
variation H can be described by
^^-_2^f(pL-m (1)
where Ay is the thickness variations of the skull
bone; Cs is the velocity of sound in the skull
210
1
TRANSDUCER SKULL
AY
Fig. 2. Phase aberrations introduced by the
presence of skul 1 .
bone and Cg is the velocity of sound in brain.
Based on the attenuation data and skull thick-
ness variation, the skull bone can be modeled as
an attenuating plate of varying thickness posi-
tioned in front of the transducer array. The
validity of this model has been demonstrated by
studies of acoustic transmission through skull
flaps performed in a water tank environment with
a linear array transducer (Phillips et al . [8]).
Acoustic field plots, experimentally measured
through skull, were compared with calculated
field plots using thickness data from the same
skulls. If the proposed model of the skull is
valid as a first order approximation, then the
character of the calculated plots should corre-
late closely with the experimental field measure-
ments. Sections of skull from the temporal re-
gion were frozen upon removal , thawed before
experimental use and then continuously stored in
water. A diagram of the experimental arrangement
is shown in figure 3. The multielement array
measured 24 mm in length and consists of 16,
equally-spaced, transducer elements. Each ele-
ment is 0.35 mm in width and 14 mm in elevation.
The center frequency was experimentally measured
to be 1.8 MHz in water. The linear array was
placed just under the surface of the water with
the skull flap positioned in close proximity.
Independent control of the transmit timing delay
for each element was provided by continuously
variable, mono-stable multivibrators whose out-
puts were directly coupled into the solid state
transmitter pulsers. The acoustic field patterns
of the array were recorded using a small trans-
ducer probe which could resolve small spatial
variations of acoustic pressure. A Helix trans-
ducer probe measuring 0.46 mm in diameter was
used for this purpose. The acoustic pressure
was displayed on a cathode ray tube and recorded
on photographic film as a function of the later-
al position of the translated field probe.
The Helix probe was then moved aside and a
simple piston transducer 12.5 mm in diameter was
used as an illuminating transducer which provided
TRANSDUCER
Fig. 3. Measurement system for transducer field
plots .
a smooth acoustic wavefront incident at the inner
table of the skull. The relative arrival times
of the acoustic energy were recorded at the in-
dividual array elements. The arrival time or
phase variations are directly related to the
variations in thickness of the skull across the
transducer aperture. As a theoretical check of
the proposed model of skull bone, computer simu-
lations of acoustic field pressure were con-
structed using the relative phase shifts of the
acoustic pulses which had traveled through the
skull to the individual array elements.
Figure 4 compares the experimental and cal-
culated acoustic field plots which are used to
evaluate the ultrasound beam character. Relative
acoustic pressure was normalized to the spatial
maximum of each scan and plotted as a function
of lateral position. The extent of the lateral
translation is 25 mm. The element transmit de-
lays were set for a 10 cm Gaussian focus, and
the experimental field plot was recorded in the
focal plane.
The upper plot shows the measured acoustic
pressure at a range of 10 cm without the presence
of a skull flap; the middle plot shows the meas-
ured acoustic pressure when the adult skull is
placed in front of the linear array. Of parti-
cular interest is an azimuthal shift of the main
lobe to the right and the broad main lobe beam-
width. The bottom plot shows the computer cal-
culation where the experimentally measured ar-
rival times were incorporated into the transmit
timing sequence to simulate the acoustic field
pattern. In each simulation a transmit pulse
character was specified along with an attenua-
tion factor for each element to more accurately
represent the experimental situation. However,
the amplitude variations across the transducer
aperture appeared to be of little significance
for thickness variations on the order of ultra-
sonic wavelengths. If the proposed model of the
skull is valid as a first order approximation,
then the character of the simulation should cor-
relate closely with the acoustic field plot
obtained experimentally. The computer simula-
tion shows an azimuthal shift of 4.9 mm to the
211
0
1 1 —
10 0
1 i 8
1 1
f u
1
/I
1 1
ft 1 9
0 1 L
UJ
1.0
THROUGH
'LITUC
.75
. SKULL
2:
.50
UJ
>
1—
■a:
_]
UJ
.25
0
1 1
1 I
1
1 1
12 8
4 0
4
8 12
PHASE
COMPENSATION
Fig. 4. Comparison of experimental versus cal-
culated transducer field plots through
the skull.
right compared with the 5.0 mm shift in the
experimental plot. The 6 dB beamwidths in both
plots are noted to be 7.0 mm. The overall geo-
metrical similarities lend considerable support
to the proposed first order model of the skull.
4. Resolution Limitations in the
Presence of Phase Variations
One should now consider what is the extent of
image degradation due to skull thickness varia-
tion in B-mode echoencephal ography . As has been
briefly discussed in a previous publication
(Smith et al. [7]), the skull thickness itself
and hence the phase across a one-dimensional
transducer can be written as an
polynomial :
,th
order
A +
C2
x
D3 + E"^
X x
(2)
where <^ is the phase variation and x is the loca-
tion on the skull or transducer. Thus the phase
variation due to the skull can be described as
the sum of a mean phase shift, a linear phase
variation and higher order terms. On the average
the magnitude of the coefficients will decrease
with higher orders, i.e., the finer grain varia-
tions will have smaller amplitudes than, for in-
stance, the linear thickness variation of the
skull across the transducer. It is also evident
that the thickness variations of the skull will
have less significance for imaging systems using
longer wavelengths.
The mean thickness and the 1st order thickness
variation of the skull make up the first two
terms of the phase polynomial. The mean thick-
ness of the skull with its acoustic velocity of
3050 m/s results in a range shift in an individ-
ual A-mode or B-mode line. A linear thickness
variation of the skull results in a refraction
of an A-mode or B-mode line according to Snell 's
law. A range shift or an azimuthal shift does
not alter the beam width of a transducer. Hence,
these effects do not degrade the resolution of an
A-mode line or individual line of a B-mode image.
However, for a linear B-scan or a compound B-scan,
a linear skull phase variation which changes as
a transducer is moved across the skull can cause
distortions and misregistration in both the
range and lateral dimensions. A phased array
sector scanner as described by Somer [11] or
Thurstone and von Ramm [12] produces a B-scan
image in real time through a single fixed spot
on the skull. Targets comprising such a sector
image are shifted uniformly in range and azimuth
by the first two terms of the phase polynomial
and consequently there is no image distortion.
Second and higher order skull phase variations
across the transducer aperture act like an unde-
sired lens and degrade the lateral resolution of
a single A-mode line and every type of B-mode
scanner. The more random the phase variations,
the more disrupted is the diffraction pattern of
the transducer. In fact, an upper limit can be
put on the resolution capability of conventional
imaging through a phase aberrating medium (Good-
man [13]). If A is the average linear dimension
of the transducer over which the phase aberra-
tion is constant to within, say, one radian,
then the resolution capability of the system is
approximately that of a diffraction limited sys-
tem with an effective transducer aperture, Dg^^,
related to A by
D
eff
A
(3)
where F is the distance from the transducer to
its focal point, and p is the distance from the
transducer to the aberrating medium. Transducer
apertures of size greater than Deff collect more
energy but have no greater resolution capability.
Note that the most severe resolution limitation
occurs for p - 0. that is, when the perturbing
medium is directly adjacent to the transducer
in which case Dgff = A. Thus, a contact scanner,
no matter what its aperture size, is limited to
the same resolution as a transducer of size A.
Long path, focussed, mechanical water bath scan-
ners such as have been described by Fry et al .
[1] and Thurstone et al . [14] can maintain
larger effective apertures, since the trans-
ducer is sufficiently removed from the head so
that the surface area of the skull subtended by
the beam is very small and the relative phase
variation is less significant. However, these
mechanical scanners must move over the surface
of the skull making them susceptible to reg-
istration errors due to refraction effects
212
from linear phase variations as explained above.
Tomograms from a contact electronic sector scan-
ner suffer no distortions from linear phase
variations. Therefore, if much of the thick-
ness variation of the skull is a linear varia-
tion, an electronic sector scanner will be able
to maintain much of its normal resolution
capabi 1 ity .
One should now consider how large the thick-
ness variation or phase variation is in the
skull in the areas most commonly used for
acoustic windows.
For a conventional imaging system, the amount
of deviation of <fi(x) from the constant value A,
in eq. (2) , wi 1 1 be a good predictor of how badly
the image resolution will be degraded by phase
aberrations .
Fried [15] has considered this question for
astronomical imaging, wherein the changing re-
fractive index of the moving air causes a phase
aberration affecting optical telescopes. One
function which can be used to predict resolution
limitation is the root mean square phase varia-
tion, normalized to aperture size plotted as a
function of aperture size.
D/2
/ [
<j;(x) - ^ury
-D/2
dx
(4)
< > is an ensemble average, T(xT is the mean phase
across the aperture, and (()(x) is the phase shift
due to the skull at some point, x, within the
aperture. To examine this function, phase varia-
tions were measured for the 16 elements of a
linear array which are separated by 1.5 mm. Sets
of readings were taken at 10 different positions
in the temporal regions from an adult skull and a
pediatric skull. The measurements were made by
recording the time of flight of an acoustic pulse
as it propagates through the skull and arrives at
the elements of the linear array.
For a given aperture size, the mean phase was
first determined, and then the squares of the de-
viations were calculated for each element within
that aperture. The sum of the squares was taken
and then normalized to the aperture size. This
was done for every possible aperture of elements
within one set of measurements e.g. , for an aper-
ture size of eight elements there are nine pos-
sible apertures within each 16 element array.
The procedure was performed for each of the ten
sets of measurements; a final ensemble average
was made, and the square root was taken.
Figure 5 shows the result of the calculations
for two frequencies. The root mean square phase
variation is plotted in terms of radians versus
aperture size for 1.8 MHz and 1 MHz. The func-
tion is monotonically increasing at least up to
an aperture size of 24 mm. It will level off at
some larger size since the maximum thickness
variations of the bone are not more than a couple
of millimeters.
Fried [16] chooses the criterion that for root
mean square (rms) phase variations of 1 radian
(a/2tt), the resolution capability has reached an
upper limit no matter what the aperture size.
Increasing the aperture dimension beyond this
point, increases the phase variation so rapidly
that many elements on the aperture are always out
of phase. From figure 5, the phase variation is
8 16 24
APERTURE SIZE (mm)
Fig. 5. Root mean square skull phase deviation
from mean phase shift versus aperture
size.
one radian for an aperture of 6 mm at 1.8 MHz.
Therefore, the maximum effective aperture for a
conventional contact ultrasonic scanner at 1.8
MHz is 6 mm, i.e., a numerical aperture of 7.3
wavelengths (ratio of aperture size to wave-
length) and a resolution capability of 7.8 de-
grees using the Rayleigh criterion. At 1 MHz
the rms phase variation reaches one radian for
an aperture size of 18 mm. Therefore, the maxi-
mum effective aperture for a conventional con-
tact ultrasonic scanner at 1 MHz is 18 mm or a
numerical aperture of 12.2 wavelengths and a re-
solution capability of 4.6 degrees. One con-
cludes then that in the presence of such skull
variations a conventional contact B-scanner can
achieve a larger useful numerical aperture and
hence better lateral resolution at 1 MHz than
at 1.8 MHz.
At this point it would be interesting to make
another set of calculations which would be ap-
plicable to an electronic sector scanner using
a contact transducer. It has been mentioned that
such an imaging system is insensitive to linear
phase variations across its aperture. Conse-
quently, the relevant function to be calculated
is the rms phase deviation, R^, from the phase
slope for a given aperture size; i.e., the de-
viation from
<^l{x) = A + Bx
(5)
Following the same line of thought as above, we
now calculate
D/2
-D/2
(t)(x) - (f>|^(x)
dx > ,
(6)
where (j)|_(x) is the slope of the phase data for a
given aperture size. In this series of calcula-
tions, using the same data as above, the slope
and intercept were determined by the least squares
methods, and the calculations proceeded as above.
Figure 6 shows the results to be as expected.
For an electronic sector scanner at 1.8 MHz, the
rms phase variation reaches a critical value of
213
LU
a.
o
Ui 1.5r
APERTURE SIZE (mm)
Fig. 6. Root mean square skull phase deviation
from phase slope versus aperture size.
one radian at an aperture size of approximately
14 mm. Therefore, on the average, the maximum
effective aperture size will be 14 mm or 17 wave-
lengths and the resolution capability will be
3.4 degrees.
Now at 1 MHz, the rms phase variation for a
sector scanner never reaches one radian for the
Fig. 7. Horizontal cross sectional ultrasound
image of brain through adult skull com-
pared to anatomical brain section.
aperture data which we measured. If we extrap-
olate this curve, we find that it will reach one
radian near 1.25 inches or 32 mm. This cor-
responds to a maximum effective numerical aper-
ture of 21 wavelengths and a resolution capa-
bility of 2.7 degrees.
These calculations, based on limited data,
have yielded consistent results. For the one-
dimensional skull data analyzed here, a 1 MHz
transducer can achieve a larger effective numer-
ical aperture and hence better lateral resolu-
tion than can a transducer operating at a higher
frequency. The considerations of the increasing
attenuation with frequency of ultrasound by the
skull lend added support to the choice of a
1 MHz transducer for transkull imaging. In ad-
dition since a phased array sector scanner is
insensitive to linear phase variations due to
the skull, such a system would offer advantages
in maintaining its resolution capability over
conventional B-scanners.
A representative image is shown in figure 7
using the real time phased array B-scan system
with a 1 MHz, 31 mm linear array transducer.
A horizontal ultrasound tomogram through the
skull of a 45 year old female is compared to an
anatomical cross-section (Roberts and Hanaway
[17]). As indicated, one can see the far skull,
the mid-cerebral fissure, the posterior areas
of the lateral ventricles, mid-line structures
and echoes from the sylvian fissure. The corpus
callosum appears as the relatively echo free
band between the mid-cerebral fissure and the
posterior areas of the lateral ventricles.
Images of similar quality along with real time
anterior horns of
lateral ventricles
Fig. 8. Coronal cross sectional ultrasound image
of brain through skull compared to ana-
tomical brain section.
214
display or pulsating cephalic blood vessels have
been obtained in both horizontal and coronal
sections .
Figure 8 shows a coronal scan in a ten year
old normal female taken slightly posterior to
the pinna of the ear. In this view the far
skull, the mid-cerebral fissure, the anterior
horns of the lateral ventricles, the third
ventricle and the near side sylvian fissure are
visualized. Furthermore, the corpus callosum
is again seen as a relatively echo-free band
above the anterior portions of the lateral
ventricles. In this view the posterior cerebral
arteries are consistently seen as pulsating
echoes in the lower portion of the scan on both
sides of the mid-line. In addition, the branch
of the middle cerebral artery is also seen quite
frequently in the area of the sylvian fissure.
5. Resolution Improvement via
Phase Compensation
Having arrived at an improved transducer con-
figuration to overcome partially the attenuation
and phase aberration effects of the skull, signal
processing techniques have been used in prelimi-
nary experiments to remove more completely the
effects of skull thickness variations in a phased
array imaging system.
Perhaps the most logical signal processing
method is to measure the phase variation <i>{x) of
the aberrating-medium before an image is made,
determine a compensating phase variation, -<t>(x),
and then add such a phase compensation to the
imaging system in the form of an acoustic lens.
Independent control over each element of the
transducer array in transmit and receive modes
provided a method to produce any desired wave-
front response within the limitations imposed by
element size and spacing with respect to the
wavelength of ultrasound used. Such a process
has long been recommended to compensate for lens
aberrations in optical telescopes (Tsujiuchi
[18]).
Recently several groups have reported con-
struction of large sampled aperture telescopes
whose mirror elements will move to compensate
for the time varying phase aberrations of the
turbulent atmosphere in real time (Muller and
Buffington [19]; Hardy et a1 . [20]).
Figure 9 illustrates the basic principle of
the phase compensation technique in an elec-
tronic phased array scanner (Phillips et al .
[8]). Five representative elements of the trans-
ducer array are chosen with a section of skull
placed in front of them. Part A shows the phase
aberrated wavefronts emerging from a variable
thickness skull when the elements are phased to
produce a focussed wavefront in a medium of
constant acoustic velocity. In B, an acoustic
wavefront provided by another source passes from
the inner table of the skull to the transducer
elements. Knowledge of the spatial character
of the incident wavefront and the arrival times
at each transducer element allows for the de-
termination of the relative changes in skull
thickness in front of the elements comprising
the aperture. When these relative phase varia-
tions are incorporated into the transmit timing
as shown in C, the emergent wavefronts exhibit
a restored phase character similar to that
originally intended.
received echo programmable
Fig. 9. Principle of the phase compensation
technique.
The following experiment using the equipment
configuration of figure 3 is presented to il-
lustrate preliminary findings (Phillips et al .
[8]) utilizing phase compensation. As shown in
figure 10a the control field plot was recorded
at a range of 10 cm when the transducer elements
were phased for a 10 cm focus and the skull re-
moved. The 6 dB beamwidth was experimentally
measured to be 4.5 mm compared with a calculated
4.3 beamwidth for the diffraction limited aper-
ture. Figure 10b shows the beam plot with the
adult skull positioned between the array and
field probe. The 6 dB beamwidth was increased
to 14.0 mm, and a main lobe refraction of 3.0 mm
to the right is also apparent. The relative
phase changes as a function of skull position
in front of the linear array were measured, and
the phase compensated beam profile is shown in
figure 10c. Although the acoustic beam is at-
tenuated due to a single transmission through
skull bone, the phase character due to variations
in skull thickness is restored. The refracted
lobe was returned to within 0.5 mm of the axis
defined in the control, and the 6 dB beamwidth
was significantly improved and measured to be
4.5 cm. Although the foregoing studies were
performed for the transmit mode in a direction
normal to the face of the transducer, it was
hoped that similar improvements could be real-
ized when the ultrasound pulse was electroni-
cally steered and focussed in the receive mode.
The next step was to perform phase com-
pensation studies for the transmit-receive
mode in the water tank environment through a
215
mm
Fig. 10. Acoustic field plots demonstrating phase
compensation in the transmit mode.
preserved skull segment. A phased array B-scan
system capable of 2 to 4 mm resolution in range
and azimuth throughout a 15 cm field of view
was used (von Ramm and Thurstone [3]). A 31 mm,
1 MHz, linear array was used to minimize the ef-
fects of skull attenuation and reduce the rela-
tive phase variation across the transducer as has
been described above. To implement the phase
compensation technique, pulses from the array
passed through the skull and were reflected by a
wire target located at a range of 7 cm. Not-
withstanding the aberrated character of the ini-
tial transmitted pulse, the echoes from the wire
target return to the skull as spherical waves
since the wire approximates a line reflector of
infinitesimal size.
Figure 11 illustrates the process of phase
compensation. Part A in figure 11 shows the re-
ceived echoes for five representative channels.
The returning echoes arrive at individual ele-
ments of the array according to the spherical
character of the incident wavefront. Since the
programmable delay lines are sequenced to pro-
vide a focus for all points throughout the ob-
ject volume of interest, the electronic pulses
corresponding to the received echoes emerge from
the delay lines at identical times as shown to
the right. In part B a skull sample was placed
in front of the linear array. Due to the aber-
rating nature of the skull the emerging wavefront
no longer resembles a diverging spherical wave.
Since the delay times are sequenced for a focus
in a homogeneous medium it is not surprising that
Fig. 11. Principle of phase compensation tech-
nique in the receive mode.
the outputs from the delay lines are unable to
provide phase coherence for an echo returning
from a point target when the skull is interposed.
In part C the received echoes from the same point
target through the skull flap are restored to
phase coherence by adjusting the delays of each
delay line under software control, so that all
outputs arrive in time coincidence with that of
one channel chosen as reference. This procedure
removes the timing errors caused when imaging
through skull of varying thickness or composition.
Results are shown in the images of figure 12.
Each image was made with the gain and reject
controls set for optimum resolution. The left
hand control image shows the ultrasound image of
a line target used to produce the illuminating
wavefront and a series of strings of variable
spacing (5 mm to 10 mm) without the presence of
a skull. The center image shows clearly the de-
graded resolution in the image of the same targets
through a section of bone from the temporal area
of the skull. The right hand image shows a sig-
nificant restoration of image resolution for the
line target and the strings in the center of the
field of view. Decreased off -axis sensitivity
216
CONTROL
THROUGH SKULL
THROUGH SKULL WITH
PHASE COMPENSATION
Fig. 12. Resolution targets through skull using phase compensation.
through the skull as the critical angle was ap-
proached was responsible for the target dropout
at the edges of the field of view. The phase
compensation technique clearly reduced the spot
size of the targets imaged through the skull,
thus restoring much of the azimuthal resolution
of the system.
Implementation of the phase compensation tech-
nique in a clinical environment will be more com-
plex since additional procedures and techniques
must be utilized for ease of clinical use. Of
basic importance is the necessity of producing
a smooth acoustic wavefront incident at the inner
table, within a live, intact head. One can specu-
late that the pineal gland which is calcified in
most adults or the relatively planar structures
of the third ventricle or the far side of the
skull may serve as adequate reflectors to use to
measure the phase variation of the inner table
of the skul 1 .
The results presented here are still prelimi-
nary. It is difficult to predict how much im-
provement will be necessary in lateral resolu-
tion and target dynamic range before B-mode images
of cephalic structures routinely provide informa-
tion of diagnostic value. However, the improve-
ments presented here are encouraging and may lead
to new techniques of acoustic imaging of cephalic
structures having acceptable resolution for diag-
nostic evaluation.
Acknowledgments
This work was supported by USPHS grants
HL-12715, HL-14228, HS-01613, and by the Food
and Drug Administration, Bureau of Radiological
Health.
References
[1] Fry, F. J., Eggleton, R. C. , and Heim-
burger, R. F., Transkull Visualization
of Brain Using Ultrasound: An Experi-
mental Study, in Ultrasonics in Medicine,
M. de Vlieger, D. N. White, and V. R.
McCready, eds., pp. 97-103 (American
Elsevier Publishing Company, Inc., New
York, 1974).
[2] Brinker, R. A. and Taveras, J. M., Ultra-
sound cross-sectional pictures of the head,
Acta. Radiol. Diag. 5, 745-753 (1966).
[3] Von Ramm, 0. T. and Thurstone, F. L.,
Cardiac imaging using a phased array ultra-
sound system I: system design. Circulation
53 (2), 258-262 (1976).
[ 4 ] Wh i te , D . N . , Ultrasonic Echo-encephal o-
graphy . Medical Ultrasonic Laboratory,
Kingston, Ontario, Queen's University, 1970.
[5] Hueter, T. F., Messung der Ul traschal 1 ab-
sorption in menschlichen Schadel knochen und
ihre Abhangigkeit von der Frequenz, Natur-
wissenschaften 39, 21-22 (1952).
[6] Fry, F. J., Barger, J. E., and Sanghivi,
N. T., Acoustic Properties of Human Skull
Bone. (Proceedings of the World Federation
of Ultrasound in Medicine and Biology), in
Ultrasound in Medicine, D. N. White and R.
Brown, eds.. Vol. 3, p. 2095 (Plenum Press,
New York, 1977).
[7] Smith, S. W. , Phillips, D. J., von Ramm,
0. T., and Thurstone, F. L., Real Time B-
mode Echoencephalography , in Ultrasound in
Medicine, D. N. White, ed.. Vol. II, pp.
373-382 (Plenum Press, New York, 1976).
[8] Phillips, D. J., Smith, S. W., von Ramm,
0. T., and Thurstone, F. L. , Sampled Aper-
ture Techniques Applied to B-mode Echo-
encephalography, in Acoustical Holography,
N. Booth, ed.. Vol. 6, pp. 106-119 (Plenum
Press, New York, 1975).
[9] Martin, B. and McElhaney, J. N. , The
Acoustic properties of human skull bone,
J. Biomed Mater. Res. 5, 325-333 (1971).
[10] Wells, P. N. T., Physical Principles of
Ultrasonic Diagnosis (Academic Press, New
York, 1969).
[11] Somer, J. C, Electronic sector scanning
for ultrasonic diagnosis. Ultrasonics 6,
153-159 (1968).
[12] Thurstone, F. L. and von Ramm, 0. T. A
New Ultrasound Imaging Technique Employing
Two-Dimensional Electronic Beam Steering,
in Acoustical Holography, P. S. Green, ed..
Vol. 5, pp. 249-250 (Plenum Press, New York,
1974).
217
[13] Goodman, J. W. , Huntley, W. H. Jr., Jackson,
D. W., and Lehman, M., Wavefront-Reconstruc-
tion Imaging Through Random Media, Appl ied
Physics Lett. 8, 311-313 (1966).
[14] Thurstone, F. L., Kjosnes, N. I., and Mc-
Kinney, W. M., Ultrasonic scanning of bio-
logic tissue by a new technique, Science
149,-302-303 (1965).
[15] Fried, D. L., Optical resolution through a
randomly i nhomogeneous medium for very long
and very short exposures, J. Opt. Soc.
Amer. 56, 1372-1379 (1966).
[16] Fried, D. L., Statistics of a geometric
representation of a wavefront distortion,
J. Opt. Soc. Amer. 55, 1427-1435 (1965).
[17] Roberts, M. and Hanaway, J., Atlas of the
Human Brain in Section, pp. 22-47 (Lea and
Feiberger, Philadelphia, 1970).
[18] Tsujiuchi, J., Correction of Optical Images
of Compensation of Aberrations and bv Spa-
tial Frequency Filtering, in Progress in
Optics, II, E. Wolf, ed., pp. 133-182
(John Wiley and Sons, New York, 1963).
[19] Muller, R. A. and Buffington, A., Real time
correction of atmospherically degraded
telescope images through image sharpening,
J. Opt. Soc. Amer. 64, 1200-1210 (1974).
[20] Hardy, J. W., Feibleib, J., and Wyant,
J. C, Real Time Phase Correction of
Optical Imaging Systems, in Digest of
Optical Propagation Through Turbulence
(abstract only). July 9-11, 1974, Boul der ,
Colorado, ThBl-l-ThBl-4.
218
Chapter 8
IMAGE RECONSTRUCTION
J
219
Reprinted from Ultrasonic! Tissue Characterization II, M. Linzer, ed.. National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
CHARACTERIZATION OF IN VIVO BREAST TISSUE BY ULTRASONIC
TIME-OF-FLIGHT COMPUTED TOMOGRAPHY
G. H. Glover
General Electric Company
Medical Systems Division
Applied Science and Diagnostic Imaging Lab
Milwaukee, Wisconsin 53201, U.S.A.
The use of ultrasonic time-of-f 1 ight (TOF) computed tomography for characteriza-
tion of tissue in live breasts is reported. Quantitative distributions of the re-
fractive index within a tomogram of the specimen were obtained by reconstruction from
5 MHz TOF projection data. Fifteen breast cancer patients and five asymptomatic
volunteers were scanned during the clinical feasibility study. The results indicate
that young, dense breasts have wide variation in the refractive index distributions.
In older subjects, however, various lesions are found to have distinctive indices.
Histograms of the reconstructions show differences between pathological and normal
breasts.
Keywords: Breast cancer; computerized tomography; mammography; time-of-f 1 ight;
tissue characterization; ultrasonic imaging; ultrasound.
1. Introduction
Ultrasonic energy potentially provides an at-
tractive modality for mass screening programs
in breast cancer detection. Its advantage is
particularly evident in light of the recent
publicity concerning suspected hazards of radio-
graphic procedures. It is, therefore, of in-
terest to characterize quantitatively the ultra-
sonic properties of breast tissue, with an eye
towards differentiation of various neoplasms and
breast parenchyma. A useful vehicle for this
purpose (and perhaps for mass screening as well)
is computerized time-of-f 1 i ght (TOF) tomography.
TOF tomography was first published by Green-
leaf et al. [l]i in 1975. Since then, several
other workers [2,3] have published reports on
other aspects of this technology. In this tech-
nique, the speed of propagation of an ultrasonic
wave is accurately measured along many uniformly
sampled paths through the specimen (see fig. 1).
The data are obtained from measurements of the
time delay for reception of a short pulse travers-
ing the path. In the parallel-scan geometry
shown in figure 1, the set of values corresponding
to a given angle relative to the specimen collec-
tively form one TOF projection (or view). Other
projections are then obtained by rotating the
scanner relative to the specimen. A two-
dimensional refractive index distribution is then
obtained from the projection data by computerized
image reconstruction techniques [4]. The result-
^Figures in brackets indicate literature
references at the end of this paper.
SIDE VIEW
Fig. 1. Geometry of scanner. A single pair of
transducers are translated in 1 mm steps
to form one projection. One hundred
projections are obtained by rotation of
the scanner through 180 degrees. Other
slices are acquired with the scanner
lowered relative to the breast (side view).
221
ing images are displayed as an array of square
pixels, each of which represents the local speed
of propagation supported by the tissue. The
process is very much analogous to x-ray computed
tomography methods.
The speed of propagation of an ultrasound wave
depends upon the elasticity and mass density of
the propagating medium. Thus, the TOF reconstruc-
tions directly provide quantitative, localized in-
formation about significant mechanical properties
of the bulk tissue. (By contrast, echo amplitude
is a very complicated function of interfacial and
bulk tissue ultrasonic characteristics, as well
as of geometry.) The propagation velocity has
been found to vary by several percent in various
types of soft tissue [5]. In this paper, velocity
measurements of tissue structure in in vivo breasts
are presented and analyzed with the aid of histo-
grams. In the following section, the experimental
technique is described.
2. Experiment
A. Technique
Only a brief description is given here as the
details have been presented elsewhere [3,6]. The
scanner used in the study is shown in figure 2. A
transmitter and a receiver transducer (Panametrics ,
5 MHz) were rigidly mounted on a carriage which
could be translated by a stepping motor. Samples
were acquired at 1 mm intervals (163 samples/
projection). The tank was mounted under a canvas
sling supported by a metal and wooden structure so
that the subject lay prone with the breast im-
mersed in the water of the inner tank. Thus, hori-
zontal slices 5 mm thick transaxial to the breast
were obtained as shown in figure 1. After the
first projection was acquired, additional projec-
tions were measured as the tank was rotated through
180 degrees in 1.8 degree steps by a second motor.
The TOF data (n seconds of delay) were stored on
digital cassette tape for later reconstruction.
Fig. 2. Scanner mechanism mounted in wooden frame.
Both tanks are water filled. The breast
is immersed in inner tank, which is isolat-
ed for electrical precaution.
The reconstructions were obtained using the con-
volution algorithm [4]. The original reconstruc-
tion utilized time-sharing software and were dis-
played in a 57 x 57 matrix format [61. For the
present work, the images were re-reconstructed in
a 64 x 64 matrix representing 12 x 12 cm with a
minicomputer. Reconstruction time was about 60
seconds using conventional FORTRAN coding.
The velocity values v are given in CTU values,
defined as [6]
CTU
VO
'0
x 1000,
(1)
where vg is the velocity of propagation in the
water (at about 34 °C). Thus, +10 CTU corresponds
to a velocity 1 percent greater than for water.
The clinical study was performed at the Albany
Medical College. Fifteen symptomatic patients and
five asymptomatic volunteers were scanned. Mam-
mograms and clinical reports for the patients were
available before the scans to help localize the
region of interest. Generally, three slices 1 cm
apart for the pathological breast and one slice of
the contralateral breast were acquired. After the
TOF scans, biopsies were performed on the suspi-
cious breasts. These results were then made avail-
able for comparison with the reconstructions. A
thorough historical comparison would have been
desirable, but was outside the scope of this study.
B. Results
Figures 3 and 4 show typical reconstructions of
breasts having lesions. In these pictures, high
velocities are depicted as black. Figure 3 shows
,a right breast (age 53) containing a large (be-
nign) fibroadenoma. The CTU values for the lesion
are 20-28, while the mammory fat (white areas) has
CTU values of about -60. The left breast depicted
on figure 4 (age 44) contains infiltrating ductal
carcinoma with CTU numbers 40-45. The mammory fat
has CTU values similar to those in figure 3.
The streaking artifacts in these pictures de-
rive from patient motion during the long scan time
required to obtain the data (about 9 minutes/
slice). As a result, the pictures are much noisier
(standard deviation, 10 CTU in the water bath)
than the intrinsic velocity resolution of the
system determined with a stationary phantom (a
2 CTU). Nevertheless, the reconstructions are of
sufficient accuracy to characterize the large-mass
lesions studied here.
Table 1 presents a summary of the CTU values for
several types of lesions as well as fat and asymp-
Table 1. Summary of CTU values for several types
of lesions, fat and asymptomatic breast
parenchyma .
Symptomatic patients (ages 33 to 85)
mammary fat
normal parenchyma
f i broadenoma/f ibrocystic disease
(benign )
carcinoma
inflammatory skin thickening
Asymptomatic subjects (ages 24 to 29)
fibrotic mammary glands, ducts
CTU numbers
-50 to -70
0 to 15
12 to 30
40 to 60
40 to 65
8 to 60
222
Fig. 3. Reconstruction of breast with large
fibroadenoma (black region).
tomatic (presumed normal) breast parenchyma. In
the older patients, the CTU values were found to
be distinctive for fat, normal parenchyma, fibrotic
tissue, and carcinoma. Malignant lesions and in-
flammatory skin thickening due to the presence of
an internal malignant neoplasm have characteristic
CTU numbers greater than 40, while benign lesions
(fibrocystic disease, fibroadenoma) have CTU values
less than 30. However, the CTU values for the
dense breasts of the younger asymptomatic subjects
span the entire range. It is, therefore, impos-
sible to differentiate solely on the basis of CTU
value neoplastic tissue from normal fibrotic
parenchyma in such cases.
In several slices wherein the lesion did not
completely fill the 5 mm axial slice thickness, the
CTU values were less than the bulk numbers in the
table. These partial volume effects arise because
the component of transmitted signal which arrives
earlier due to traversal through the high velocity
lesion ultimately becomes too low in amplitude to
be detected relative to the signal components which
exclude the lesion. The transition is not perfect-
ly sharp due to refraction and tissue dependent
attenuation effects. Excluding such cases, however,
there were no counter-examples to the CTU values
classifications in the table.
3. Statistical Analysis
A rudimentary statistical analysis of the re-
constructions was performed by computing the
histograms and cumulative distribution functions.
For this purpose, a program was developed which
searched the matrix for the edge of the breast,
fit an arbitrary ellipse to the boundary points
by regression, and calculated the histogram for
those elements within the ellipse. The class
interval was chosen as 3 CTU numbers in com-
puting the histograms. The results were normal-
ized to remove variations in specimen size.
Figure 5 shows histograms for a young, asymp-
Fig. 4. Reconstruction showing infiltrating duc-
tal carcinoma (lower quadrant) and in-
flammatory carcinoma central region.
tomatic subject. Note that the two normal
breasts have similar tissue distributions. This
was not always the case, although generally the
skew and kurtosis of the histograms were similar.
The distributions are peaked near CTU = 0 and
contain considerable fibrotic tissue with high
CTU numbers. The variations in the histograms
for different young subjects were quite wide.
This is illustrated by plotting the (inverse)
cumulative probability function.
0j = E fi (2)
-80 -60 -40 -20 0 20 40 60
CTU VALUES
Fig. 5. Histograms of right (solid) and left
(dashed) breasts of 24 year old asympto-
matic subject. The distributions are
very similar and centered near CTU = 0.
Considerable fibrotic structure is evident
from the peak position and area under the
high valued tails.
223
where f-; is the frequency for the i^'^ CTU value.
Curves for three of the subjects are shown in
figure 6. As may be seen, the curves span a large
range of CTU values.
10°
CTU VALUES
Fig. 6. Inverse cumulative probability function
for three normal, young subjects. The
zero asymptotes (which reflect peak CTU
values) show wide variation.
Figure 7 shows histograms for a 58 year old
patient with carcinoma in the right breast and a
normal left breast. Note that the peak for the
normal histogram has shifted position to low nega-
tive values relative to figure 5. This reflects
the atrophy of glandular structure (high CTU) and
CTU VALUES
Fig. 7. Histograms of 58 year old patient with
carcinoma in right breast (solid) and
normal (dashed) left breast. The normal
distribution is peaked near CTU = -40,
indicating atrophy of the glandular tis-
sue and replacement by fat. The kurtosis
and skew of the histogram for the patho-
logical breast are markedly dissimilar to
the normal one.
replacement by fat (low CTU) in the post-meno-
pausal breast. Although not enough data is avail-
able from this study, undoubtedly the peak posi-
tion is a decreasing function of age [7]. The
histogram for the pathological breast has marked-
ly dissimilar shape (skew and kurtosis) and ex-
tends to higher CTU values. The carcinoma is
contained in the high valued region. (The peak
near zero probably represents some water being in-
cluded in the histogram by the algorithm.) Figure
8 shows examples of cumulative distribution plots
-80 -60 -40 -20 0 20 40 60
CTU VALUES
Fig. 8. Inverse cumulation probability function
for a representative sampling of the older
patients. The zero asymptotic values can
be roughly correlated with tissue type.
- - - = normal, = fibroceptic/
adenoma, = carcinoma.
for older normal breasts, and breasts containing
fibrotic and carcinoma tissue. In general, figures
6 and 8 tend to reflect the results of all the
scans summarized in table 1.
4. Discussion
Human female breasts are constituted of time
varying fractions of various components. It ap-
pears plausible that each of these components
(parenchyma, glands, fat, etc. ) have ultrasonic
properties which are relatively similar between
different subjects. Indeed, the wide variation in
average and peak CTU numbers observed for young
dense breasts probably represents an incomplete
analysis of the volumetric fractions of glandular
tissue, fat, and ducts present in any given sub-
ject. In any case, a single number representing
velocity values averaged over the volume of a
breast [7] is probably too simplistic to be rep-
resentative of the breast.
The symptomatic subjects examined in this study
were chosen to have large, unambiguous lesions in
order to acquire data on the various tissue com-
ponents. These data, however, represent a statis-
tically insignificant number of cases on which to
224
base any absolute conclusions. In fact, a larger
data base will probably show wider variation in
values than was found in this study. Nevertheless,
the present results suggest a distinction between
normal and neoplastic tissue which is encouraging
for mammographic application.
There are potentially several methods of utiliz-
ing the TOF data in cancer detection suggested by
the histograms. When available, histograms of the
contralateral breast provide a reference which
could form part of a comparative basis. Alterna-
tively or additionally, a reference base which de-
pends on age and other variables can be accumulated.
Besides the histograms, which present a volumetric
picture of the constitutive elements, the individ-
ual pixel values, rates of change, and geometry can
be searched for characteristic patterns. Applica-
tion of such speculation, however, must await the
acquisition of more massive amounts of data.
Acknowledgments
The clinical data was acquired with the help of
R. W. Sponzo, M.D., his staff and colleagues at
the Albany Medical College.
References
[1] Greenleaf, J. F. et al . , Algebraic Reconstruc-
tion of Spatial Distributions of Acoustic
Velocities in Tissue from their Time of Flight
Profiles, in Acoustic Holography, W. Booth,
ed.. Vol. VI, pp. 71-89 (Pergamon Press, New
York, 1975).
[2] Carson, P. L., Oughton, T. V., and Hendee,
W. R., Ultrasonic Transaxial Tomography by
Reconstruction, in Ultrasound in Medicine,
D. White and R. Barns, eds.. Vol. II, pp.
341-350 (Plenum Press, New York, 1976).
[3] Glover, G. H. and Sharp, J. C, Reconstruc-
tion of Ultrasound Propagation Speed Distribu-
tions in Soft Tissue: Time-Of-Fl ight Tomo-
graphy, IEEE Trans. Sonics and Ultrasonics
SU-24, 229-234 (1977).
[4] Gordon, R. and Herman, G. T., Three dimen-
sional reconstruction from projections: a
review of algorithms, Internat. Rev. Cytology
38, 111-151 (1974).
[5] See, e^. , Wells, P. N. T. , Review: absorp-
tion and dispersion of ultrasound in biologi-
cal tissue. Ultrasound in Med, and Biol. 1,
369-376 (1975T: ~
[6] Glover, G. H., Computerized time-of-fl ight
ultrasonic tomography for breast examination.
Ultrasound in Med, and Biol, (in press).
[7] Kossoff, G., Fry, E. K., and Jellins, J.,
Average velocity of ultrasound in the human
female breast, J. Acoust. Soc. Am. 53, 1730-
1736 (1973). ~
225
I 11 ■ I
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed. , National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
VARIATION OF ACOUSTIC SPEED WITH TEMPERATURE IN VARIOUS EXCISED
HUMAN TISSUES STUDIED BY ULTRASOUND COMPUTERIZED TOMOGRAPHY
B. Rajagopalan, J. F. Greenleaf, P. J. Thomas,
S. A. Johnson, and R. C. Bahn
Biodynamics Research Unit
Departments of Physiology and Biophysics, and
Department of Pathology and Anatomy
Mayo Clinic and Mayo Foundation
Rochester, Minnesota 55901, U.S.A.
Variation of acoustic speed as a function of temperature was measured in fresh ex-
cised human liver, psoas muscle, spleen, spinal cord, kidney and fat, parenchyma and
muscles associated with female breasts. Tissues were encased in rubber finger cots
and suspended in a temperature controlled water bath. A prototype clinical ultrasound
breast scanner was used to obtain data required to reconstruct distributions of acous-
tic speed within two-dimensional planes through the tissue specimens over a tempera-
ture range of 20 to 40 °C. Quantitative images (printer listings of acoustic speed)
of 64 X 64 pixels were used to obtain averages of up to 16 speed measurements within
the image of each tissue. The acoustic speed in most tissues increased monotonical ly
with temperature following the behavior of water. Acoustic speed of fat showed an
anomalous decrease in acoustic speed around 34 °C suggesting possible phase transition.
Key words: Aperture synthesis; computed tomography; Doppler; fluid flow; high
resolution; reconstruction; temperature reconstruction; ultrasound.
1. Introduction
Historically, production of images by ultra-
sound scanning has been done by displaying either
the pulse echo reflected from an object (B-scan)
or by displaying the attenuated signal (C-scans)
transmitted by an object [l]i. These modalities,
although useful, give only a qualitative mapping
of tissue interfaces and geometries. Images
representing quantitative distributions of basic
mechanical tissue properties as characterized by
ultrasound scattering cross section, impedance
[2,3] and acoustic speed [4] may be intrinsically
more valuable than the qualitative images rep-
resenting tissue borders and geometries. A
quantitative imaging modality would eliminate
operator-dependent variations one often finds in
the qualitative imaging modalities. Ultrasound
computerized tomography [4] is one such imaging
modality by which the distribution of one funda-
mental mechanical parameter, i.e., acoustic
speed, in a tissue can be studied.
The purpose of this report is to describe
methods and initial results of applying computer-
ized tomography to the problem of obtaining
quantitative images representing the distribution
of acoustic speed and the effect of temperature
on these values for tissues from various parts of
^Figures in brackets indicate literature
references at the end of this paper.
the body, especially the human breast. The
knowledge of temperature coefficients of acoustic
speed in tissues may be useful in estimating the
temperature change in a region of tissue occur-
ring due to external causes such as ultrasonic
heating or drugs. Since various disease proces-
ses alter tissue temperature, the temperature co-
efficients may also be helpful in the interpreta-
tion of computerized tomography images of live
tissues.
The reconstruction problem in ultrasound com-
puterized tomography has been described else-
where [4] and will not be described in detail
here. A good introduction to the basic mathe-
matics of computerized tomography and a general
overview of the analytical and iterative algo-
rithms can be found in the review articles by
Gordon and Herman [5] and Brooks and Dichiro [6].
The convolution reconstruction algorithm for the
divergent beam geometry was first derived by
Lakshminarayanan [7] and later a mathematically
rigorous treatment of the same problem leading to
identical results was given by Herman and his co-
workers [8]. This divergent beam convolution re-
construction algorithm is used exclusively for
all the reconstruction images in this paper.
2. Experimental Arrangement
In ultrasound computerized tomography the
profile data consist of measurements of time-of-
227
flight of an acoustic pulse through the sample
along different rays or directions. The data
are obtained by divergent beam scanning with a
single pair of transducers one on each side of
the region of interest. The schematic of the
data collection geometry is shown in figure 1.
The scan angle is a and is incremented at 0.15°
intervals using a stepping motor. After the scan
the ultrasound transmitter-receiver arm is rotat-
ed to a new e position and another scan is ini-
tiated. Thus, for each e position there is one
profile of time-of-f 1 i ght data. One hundred
twenty profiles separated in view by 3° increments
are gathered over the 360° range.
transmitter locus
—
profile data
V
P(a,e2)
Jy\\/
tissue] / /
profile data
P(a,ei)
receiver loci
Fig. 1. Data collection geometry for ultrasound
tomography, ej and 63 are the angles
made by the central ray in the scan with
X-axis for two different scans. Profile
data P(a,ei) is gathered by rotating the
receiver-transmitter arm about an axis
passing through the apex of the sector and
perpendicular to the plane of the figure.
Thus, the receiver loci are arcs. The
transmitter mounted at the axis of the
scan describes a circle as profile data
from different views (e) are gathered.
The schematic of the experimental arrangement
for collecting computerized tomography data is
shown in figure 2. The system is driven by an
Interdata 7/32 minicomputer which controls the
height and rotation angle (e) of the scanner arm
and collects the time-of-f 1 ight data from the
digital output of the time-of-f 1 ight (TOF) clock
and control box. This TOF clock and control box
is interfaced to the digital I-O of the mini-
computer and upon a command from the computer
collects time-of-fl ight data, outputs the data,
steps the scanner arm to the next orientation in
the scan and repeats this process until one profile
of data are collected. At the end of the scan the
scanner arm is rotated to the new 6 position by a
command from the minicomputer and the process of
collecting profile data resumed by a trigger to
the TOF clock and control box. These time-of-
fl ight profiles representing propagation delay of
acoustic pulses along many rays through the tissue
under examination are input to the reconstruction
programs to obtain quantitative distributions of
keyboard
height
control rotate
interdata
7/32
O
CRT
camera
^
strobe
16
TOF
signal
Time-of-Fl ight
clock & control
receive
(((V^((
^rotatef
■|hei
ght|
trigger|
transmitf*— ^
scanh
Fig. 2. Ultrasound computed tomography system:
Data collection and control. Computer
controls transmi t-del ay-and position.
Time-of-f 1 ight of signal is measured with
a 16-bit 100 MHz clock and counter which
is read by the computer at up to 400 points
along a scan profile, each point separated
by 0.15°. One hundred views each contain-
ing 400 samples and each sample containing
eight channels of 16-bit time-of-f 1 ight
data can be obtained in a period of 2.5
'minutes.
acoustic speed in the transverse plane through
the tissue.
3. Reconstruction Algorithm
The mathematical steps to arrive at the recon-
struction picture from the set of profile data
are given in this section without attempting to
derive any of them. These steps known as the re-
construction algorithm are treated extensively in
the literature and the derivation for divergent
beam geometry starting from Radon transform [9]
can be found in the reference mentioned earlier
[8].
The raw data of each profile consist of time
measurements (in seconds x 10"^) at each incre-
ment along the arc of the scan. The data for the
20 or more points closest to each end of the pro-
file are the time- of-fl ight for sound through
water, and never vary by more than one unit
(10"^ s) from the lowest to the highest. The
average of the first few points is subtracted from
each point of the profile and the differences
P(j,e) represent the profile data for reconstruc-
tion. Here 6 is the angle made by the central ray
of the scan with x-axis (arbitrarily chosen) and
j = -n . . . -1 , 0, 1 , . . . +n indexes the angles of
the 2n+l individual rays of the scan with j = 0
at the center of the scan. For each rotation
angle e the profile data are reduced by the
cosines of the angles of the individual rays from
the center line.
P(j,e) modified = P*(j,e) = P(j,e) cos(j£) (i)
where e is the constant angular increment in the
scanning motion (0.15°). The modified profile
228
P* (j>9) is convolved with a kernel to give the
convolved modified profile C(k,e), i .e. ,
kmax
C(k,6) = Z P (2)
''"'^min
The kernel is given by
Ao = (3)
A-j = 0 for even i and i = 0
where D is the distance from the axis of rotation
for the scanning motion to the axis of rotation
for rotate motion and kmin and kpiax indicate the
lower and upper limits of the scan. Finally,
the back-projection is formed, along the lines
of the acoustic rays, converging at the axis of
scan motion, with the back-projected C(k,e)
weighted with inverse square of the relative
distance of the point from the axis of the scan
to distance D. Thus, for each time-of-f 1 ight
profile at an angle e the matrix f(x,y) forming
the reconstructed picture is incremented by
Af(x,y)e = I C(k,e) ^ (4)
where r = distance of the point (x,y) from the
axis of the scan. C(k,e) is interpolated if no
ray goes through the point (x,y). The sum of all
the back-projected Af(x,y)Q for all e values
yields the reconstructed image at (x,y).
f(x,y)= Z fC(k,e)Di (5)
sum over Q
When the back-projection is finished, the value
for each pixel will be the increase or decrease
in time per centimeter at that point, compared
with water. The time-of-f 1 ight for water (in the
same units, calculated from the measured time and
known distance, and verified by the known veloci-
ty of sound in water for the particular tempera-
ture) is then added to the value of each pixel,
and the reciprocal of the sum is scaled to give
the final result in meters per second. The re-
constructed digital images represented by a matrix
of 64 X 64 pixels can be displayed in gray scale
in a Ramtek display system.
4. Experimental Results and Discussion
The accuracy with which the acoustic speed can
be determined using ultrasound computerized tomo-
graphy is demonstrated in figure 3 in the high
correlation of our data with the literature value
'(10) for 2.5 percent salt solution. Thin rubber
finger cots filled with saline served as the test
object. The agreement of the literature values
with the experimental results over the entire
temperature range is within ± 0.3 percent. Since
the reconstructed image is a matrix giving the
acoustic speed in each pixel (pixel size ^ 1.5
mm^) mean and root mean square deviation of the
Fig. 3. Comparison of acoustic speed in 2.5 per-
cent salt solution (2.5 g salt in 100 cm^
of water), as determined by ultrasound
computerized tomography, with the litera-
ture values. The literature values have
an accuracy of ± 4 cm/s. Acoustic speed
values were taken as the average of 20
samples in the reconstructed image.
oUltrasound computerized tomograph data
•Millero, F. J. and Kubinski, T.,
J_. Acoust. Soc. Am. 57^, 312 (1975).
acoustic speed are easily determined by analyzing
the quantitative distribution of acoustic speed
in the image. Similar statistical analysis was
done on the reconstruction images of all the tis-
sues that were studied and reported in this paper.
Tissues were obtained and selected by a pathol-
ogist as being representative of selected tissue
types. Samples of tissue acquired were about 2
to 3 centimeter square and 5 centimeter long and
were suspended in a bath of normal saline. Ref-
erence objects such as finger cots filled with
saline of known concentrations were also suspend-
ed with the tissues. Later in the study the
tissues were packed into finger cots under normal
saline and suspended in the water bath. This
procedure simplified the tissue handling con-
siderably and the effect of finger cots on the
quantitative results was found to be negligible.
The temperature of the water bath was in-
creased in increments of about 2 °C and the tis-
sues were allowed to equilibrate with the bath
water prior to each new scan. The temperature
of the bath was controlled ±0.1 °C and experi-
ments were done over a range of approximately
22 to 42 °C.
The ultrasound computerized tomography images
for the acoustic speeds in spleen, kidney, liver,
spinal cord and psoas muscle at two different
temperatures are displayed in figure 4. Since
the two pictures are displayed in the same gray
scale any increase in the acoustic speed with
temperature is seen as an increase in the bright-
ness of the corresponding regions of the image.
The water background is brighter in the image for
33.2 °C for the same reason. Figure 5 gives a
similar picture for breast tissues at tempera-
tures 14 °C and 41 °C. Because of the very large
temperature change, the corresponding changes in
the acoustic speeds are seen strikingly in this
picture. The water background goes from dark to
bright and the image of salt solution gets bright-
er at the higher temperature. Images of the
breast fat which were brighter than the water
229
®0
1 = spleen
2 = kidney
3 = 1 i ver
4 = spinal cord
5 = psoas muscle
Fig. 4. Ultrasound computerized tomography images
of acoustic speed distribution in various
tissues, in vitro, at two different dif-
ferent temperatures. Same gray scale used
for both the pictures. Higher acoustic
speed is displayed brighter. Reconstruc-
tion size is 15 cm square. Tissues were
encased in rubber finger cots to prevent
solutes from diffusing into or out of the
tissues. Tissues were unfixed and main-
tained at 10 to 14 °C until used some 12
to 24 hours after autopsy. (Reproduced
with permission from Greenleaf, et al . ,
Proceedings of the Vth International
Conference on Information Processing in
Medical Imaging (in press).)
background at the lower temperature become darker
at the higher temperature since the acoustic speed
in the breast fat has decreased with increasing
temperature. The dark streak from fat to fat in
the picture for 41 °C is an artifact caused by
the lens effect of the cyl indrically shaped tis-
sue. The increase in the acoustic speed in
1 = 2.5 g/100 ml salt
sol ution
2 = breast fat with
parenchyma
3 = breast fat
Fig. 5. Ultrasound computerized tomography
pictures of acoustic speed distributions
in breast tissues, in vitro, at two dif-
ferent temperatures. Reconstruction size
is 15 cm square. Fat exhibits acoustic
speed higher than water at 14 °C and
lower than water at 41 °C.
parenchyma with temperature is also seen from
this picture.
Results of the statistical analysis of the re-
construction images giving the mean and standard
deviation of the acoustic speeds at various tem-
peratures in all the tissues studied are given in
tables 1 and 2. These data are graphically il-
lustrated in figures 6, 7, and 8. From figure 6
it is clear that the tissues liver, psoas muscle,
spleen, spinal cord and kidney have higher acous-
tic speeds than water and their temperature varia-
tions are similar to that of water. The breast
muscle and parenchymal tissue are about 3 per-
cent higher in acoustic speed than normal saline.
Their temperature behavior is also very similar
Table 1. Temperature variation of acoustic speed in excised human breast tissues.
Temperature
Tissue
22.5 °C
25.8 °C
27.8 °C
30.2 °C
32.2 °C
35.1 °C
38.0 °C
40.1 °C
42.5 °C
Salt finger
{2.5%)
1516.73
(2.1)b
1523.2
(2.1)
1529
(2.6)
1533.6
(2.1)
1538.4
(2.8)
1544.8
(3.1)
1549.6
(3)
1552.8
(4.2)
1556.5
(3.8)
Fat breast
1480.7
(2.5)
1466.9
(9.9)
1472.1
(9.8)
1477.4
(9.9)
1478.1
(11.4)
1436.3
(15.6)
1435.6
(18.4)
1438.7
(18.1)
1441.8
(17.5)
Fat with
parenchyma
1494.7
(4.1)
1487.5
(5.5)
1491.5
(5.5)
1497.8
(5.7)
1500.9
(6.2)
1471.3
(9.7)
1471.3
(13.1)
1474.2
(13.4)
1475.9
(12.8)
Parenchyma
1539.4
(4.5)
1545.7
(4.0)
1551
(4.6)
1558.1
(3.9)
1562.1
(4.1)
1564.5
(7.6)
1564.3
(6.4)
1569.6
(6.3)
1571.9
(6.4)
Muscle
1543. 1
(5.2)
1551.4
(6.2)
1554.2
(6.5)
1562.4
(7.1)
1565.5
(6.4)
1566.9
(4.5)
1570.7
(6.9)
1574.6
(6.7)
1579.5
(5.4)
Background
(normal saline)
1504.0
(3.2)
1512.5
(3.6)
1519.2
(2.8)
1523.7
(2.5)
1528.4
(2.4)
1535.4
(4.0)
1539.1
(2.9)
15^1.9
(3.0)
1546.4
(2.8)
fvelocity, m/s.
Standard deviation.
230
Table 2. Temperature variation of acoustic speed in selected excised human tissues.
Temperature
Tissue
17.0 °C
22.0 °C
23.5 °C
26.2 °C
30.2 °C
33.2 °C
35.2 °C
37.2 °C
39.0 °C
40.8 °
Liver
1547.0^
(2.5)b
1555.5
(1.8)
1563.1
(3.0)
1564.6
(2.8)
1571.1
(2.5)
1573.5
(1.8)
1575.3
(2.5)
1578.1
(2.9)
1580.0
(2.2)
1581.7
(1.5)
Kidney
1508.5
(4.3)
1523.8
(4.6)
1536.1
(2.1)
1536.3
(5.2)
1545.2
(2.7)
1551.4
(1.4)
1555.8
(1.8)
1560.2
(1.8)
1562.6
(1.2)
1564.3
(0.9)
Spleen
1528.0
(1.8)
1538.8
(1.8)
1544.3
(1.8)
1549.1
(1.6)
1556.4
(2.0)
1561.9
(1.7)
1546.0
(2.1)
1567.1
(2.3)
1569.3
(2.6)
1573.0
(2.1)
Psoas
muscle
1542.5
(3.0)
1459.5
(3.0)
1554.8
(1.1)
1560.2
(1.7)
1566.4
(2.2)
1571.6
(1.8)
1573.5
(1.8)
1575.6
(1.1)
1577.6
(2.1)
1580.3
(1.8)
Spinal
cord
1509.0
(4.5)
1523.0
(4.6)
1523.0
(5.3)
1526.0
(3.0)
1532.6
(3.2)
1538.0
(2.6)
1538.0
(3.5)
1542.4
(3.3)
1543.8
(3.0)
1-156.5
(2.0)
^Velocity, m/s.
Standard deviation.
Temperature, °C
Fig. 6. Variation of velocity of sound with temper-
ature in various tissues. Ultrasound com-
puterized tomography data. Values are
averages of about 15 values taken in the
visually defined center of respective tis-
sue samples within the image. Bars are
± RMSD.
to that of saline or water as seen in figure 7.
However, breast fat and parenchymatous fat have
acoustic speeds below that of normal saline and
show a complex behavior with temperature as il-
lustrated in figure 8. Initially the acoustic
speed increases slowly with temperature and around
^35 °C decreases markedly and then increases with a
'small slope of about 1 meter per second per °C.
Similar decrease but much less pronounced is seen
around 27 °C also. A plot of the variation of the
temperature coefficient versus temperature for
fat and muscle shown in figure 9 illustrates the
strikingly different temperature behavior of fat.
This complex behavior is suggestive of possible
•muscle
□parenchyma
isalt solution, 2.5 g/100 ml salt solution
Temperature, °C
Fig. 7. Variation of acoustic speed within pec-
toral is muscle and normal parenchyma of
breast. Values obtained in the manner
described in figure 6.
phase transitions just below the body tempera-
ture (37 °C). Hoyer and Nolle [11] studied the
behavior of nematic and cholesteric liquid crys-
tals near the isotropic to liquid crystal transi-
tion by measuring the propagation constants of
ultrasonic waves. Their study indicates that the
acoustic speed goes through a minimum at the
temperature of the phase transition. Dyro and
Edmonds [12] investigated ultrasonic dispersion
in cholesteryl esters and their results show that
the acoustic speed exhibits markedly different
behavior around the phase transition. Though the
fat sample in the present study is not a homo-
geneous, "clean", one-component system, the pos-
sibility of a phase transition of some kind can-
not be excluded. More detailed studies using
different physical chemistry techniques such as
differential scanning calorimetry may shed some
light into the anomalous temperature behavior of
the breast fat.
231
Temperature, °C
Fig. 8. Variation of acoustic speed in normal
breast fat. Slope of alteration in acous-
tic speed with temperature is negative for
fat in certain regions of temperature.
Vertical shift in curves is apparently due
to presence of parenchyma which has tem-
perature dependence much like water (see
fig. 7).
Alterations of temperature within various
tissues can be expected under various inflammatory
processes such as infections. It is also well _
known that certain carcinomas are associated with
a high temperature due to their increased metabo-
lism. Knowledge about the temperature variations
of acoustic speed in diseased tissues and in
normal tissues should be valuable in future
studies of acoustic speed distributions within
patients whose tissues are undergoing various
disease processes. Computerized ultrasound tomo-
graphy may provide the possibility of quantita-
tively measuring acoustic speed in breast tissues
in vivo.
Acknowledgments
The support from Dr. Earl H. Wood, Dr. Erik L.
Ritman, and the staff of the Biodynamics Research
Unit, Mayo Clinic, is appreciated. This research
was supported by Grants HL-07111, HL-00170, HL-
00060, RR-00007, HL-04664, NIH-HT-4-2904 from the
National Institutes of Health, United States
Public Health Service, and NCI-CB-64041 from the
National Cancer Institute.
References
[l] Green, P. S., Schaefer, L. F., Jones, E. D. ,
and Suarez, J. R., A New Hi gh-Performance
Ultrasonic Camera, in Acoustical Holography,
Vol. 5, pp. 493-503 (Plenum Press, New York,
1973).
[2] Jones, J. P., Impedi ography : A New Ultra-
sonic Technique for Non-Destructive Testing
and Medical Diagnosis, in Ultrasonics Inter-
national (1973), Conference Proceedings,
p. 214 (PIC Science and Technology Press ,
U.K., 1974).
[3] Kak, A. C. and Frye, F. J., Acoustic Im-
pedance Profiling: An Analytical and
Physical Model Study, in Ultrasound Tissue
-16l 1 I I I I I I I I L_
20 24 28 32 36 40
Temperature, °C
Fig. 9. Variation of temperature coefficient of
acoustic speed with temperature in pec-
toral is muscle and in fat from breast.
Negative temperature dependence of fat
in certain regions of water is visualized
by plotting slope of figure 8.
Characterization, M. Linzer, ed. , National
Bureau of Standards Spec. Pub. 453, p. 231
(U.S. Government Printing Office, Washing-
ton, D.C., 1976).
[4] Greenleaf, J. F. and Johnson, S. A., Alge-
braic Reconstruction of Spatial Distributions
of Refractive Index and Attenuation in Tis-
sues from Time-Of-Fl ight and Amplitude
Profiles, in Ultrasonic Tissue Characteriza-
tions , M. Linzer, ed. , National Bureau of
Standards Spec. Pub. 453, pp. 109-119 (U.S.
Government Printing Office, Washington,
D.C., 1976). j
[5] Gordon, R. and Herman, G. T. , Three-
dimensional reconstruction from projections:
A review of algorithms. International Re- j
view of Cytology 38, 111-151 (1974). '
I
[6] Brooks, R. A. and Di Chiro, G., Principles !
of computer assisted tomography (CAT) in
radiographic and radioisotopic imaging,
Phys. Med. Biol, n, 689-732 (1976).
[7] Lakshminarayanan, A. V., Reconstruction
from Divergent Ray Data. Department of
Computer Science, State University of New
York at Buffalo, Technical Report 92 (1975).
[8] Herman, G. T., Lakshminarayanan, A. V., and
Naparstek, A., Convolution reconstruction
techniques for divergent beams, Comput.
Biol. Med. 6, 259-271 (1976).
[9] Radon, J., Ueber die Bestimmung von Funktion 'j
durch ihre Integralwerte Langs gewisser
Mannigfaltigkeiten, Berichte Saechsische
Akademie der Wi ssenschaf ten 69, 262 (1917).
232
[10] Millero, F. J. and Kubinski, T., Speed of
sound in seawater as a function of tempera-
ture and salinity at 1 atmosphere, J. Acoust.
Soc. Am. 57, 312-319 (1975).
[11] Hoyer, W. A. and Nolle, A. W. , Behavior of
liquid crystal compounds near the isotropic-
anisotropic transition, J. Chem. Phys. 2A^,
803-811 (1956).
[12] Dyro, J. F. and Edmonds, P. D. , Ultrasonic
absorption and dispersion in cholesteryl
esters, MoT. Cryst. Liq. Cryst. 25 , 175-
193 (1974T:
233
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed. , National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
HIGH SPATIAL RESOLUTION ULTRASONIC MEASUREMENT TECHNIQUES
FOR CHARACTERIZATION OF STATIC AND MOVING TISSUES
Steven A. Johnson, James F. Greenleaf, and B. Rajagopalan
Biodynamics Research Unit
Department of Physiology and Biophysics
Mayo Foundation
Rochester, Minnesota 55901, U.S.A.
Robert C. Bahn
Department of Pathology
Mayo Clinic and Mayo Foundation
Rochester, Minnesota 55901, U.S.A.
Brent Baxter
Department of Computer Science
University of Utah
Salt Lake City, Utah 84112, U.S.A.
Douglas Christensen
Departments of Bioengineering and Electrical Engineering
University of Utah
Salt Lake City, Utah 84112, U.S.A.
Clinical and pathology-based arguments are presented for the need for higher resolu-
tion ultrasound images. The theoretical foundation and experimental characteristics
of a high resolution, sampled aperture, reflection technique termed "synthetic focus"
imaging are given. It is shown by theory and simulation that such synthetic focus
images may be corrected for attenuation and refraction effects and thereby approach
one-half wavelength resolution. The similarity between synthetic focus and seismic
migration techniques is discussed. An example of a high resolution, seismic process-
ed (i.e., migrated) image obtained from real data at medical ultrasound frequencies
is shown. Synthetic focus imaging theory is extended to moving coordinate systems
and the effect of Doppler shift effects on echo pulse shape is discussed. A general-
ized wide aperture Doppler imaging theory is presented which suggests further improve-
ments in signal-to-noise ratio, spatial resolution and flow velocity over narrow
aperture systems is possible. A new computed tomographic flow measurement and recon-
struction technique based on transmission measurements is presented. This technique
permits imaging the three flow velocity components and temperature of homogeneous
fluids in a three-dimensional domain.
Key words: Aperture synthesis; computed tomography; Doppler; fluid flow; high
resolution; reconstruction; temperature reconstruction; ultrasound.
1. Introduction
This paper presents theoretical and experi-
mental evidence that useful measurements of tis-
sue characteristics with limiting spatial re-
solutions of about one-half wavelength are pos-
sible in static and moving media (e.g. , tissues).
This paper has three objectives: First, to
present the case for striving for higher re-
solution images of tissue parameters, second,
to describe the theoretical and experimental
techniques whereby such high resolution images
may be obtained and, third, to report progress
toward obtaining such high resolution images.
235
2. The Case for Seeking Higher
Resolution Images
A. Higher Resolution Images of Static Tissues
The first goal or purpose of this paper is in
part philosophical, yet is demonstrable, and calls
attention to an important area of investigation
(improved resolution) where future research prog-
ress can be made. It can be argued that reliable
tissue characterization requires attention to at
least three independent requirements: First,
choosing an adequate set of tissue dependent acous-
tic parameters to be measured; second, measuring
the parameters with accuracy and precision in some
set of regions in the tissue; and third, making
this measurement region as small as possible and
as densely arranged or packed as possible (i.e.,
forming an image).
The third requirement is not always necessary
to gain some information on tissue type but its
inclusion improves the certainty of characteriza-
tion. Certainly, it is clear that an early diag-
nosis of a disease state (such as cancer) is en-
hanced by detection of smaller masses made pos-
sible by higher resolution and better tissue
identification. In some cases the tissue type may
well be primarily determined by the morphology and
pattern of its structure and the change in pattern
of the surrounding tissue [l,2]i. This has its
analog in microscopic pathologic techniques where
both stain (acoustic tissue characterization
parameter is its analog) and pattern are important
in tissue typing or characterization.
In the long range, the improvement of resolu-
tion may be the most important objective for
future research progress. At present, the pri-
mary attempt to detect small focal processes such
as cancers, abscesses, infarcts of less than 0.5
to 1 cm in diameter is difficult and may not be
justified economically. However, often more
fundamental than the focal lesion itself, regard-
less of its size, is the state of the tissue sur-
rounding circumscribed lesions. The pattern of
the adjacent tissue plays a crucial role in the
identification of specific diseases by supplying
the physician with information concerning the
local context of the process under consideration.
Compared to tumor nodules, the structures of
the surrounding "normal" tissues are relatively
delicate. Their patterns are essentially deter-
mined by the dimensions of the fibrous and vascu-
lar framework of an organ. It would be desirable
to image such delicate patterns associated with
the tertiary branching of major arteries of the
heart, brain, kidneys, and lungs, the bronchial
tree, the biliary ducts of the liver and the
ductular system of the breast. Since vascular
or ductal systems at the level of tertiary branch-
ing involve tubular structures approximately 0.5
to 1.0 mm in diameter, this objective fixes the
upper limit of useful picture element size at
about 0.25 to 0.5 mm. From a practical point of
view, this level of resolution would provide at
least 20 resolutable picture elements (pixels)
over an area 2.5 millimeters in diameter. Such
an area would correspond roughly to the cross-
sectional area of the lumen and portions of the
wall of a coronary artery.
^Figures in brackets indicate literature
references at the end of this paper.
In this manner, imaging studies can eventually
become focused upon pathogenesis, prevention, and
early detection of disease rather than upon the
diagnosis of gross advanced lesions.
B. The Case for Seeking High Resolution
Images of Moving Tissues
The velocity of moving tissue may be obtained
by two general methods. First, by a differential
method in which velocity is computed from images
of the location of the tissue at two separate
times separated by a short time interval (this
could be done by comparing two separate real-time
B-scans of the heart, for example), and second,
by direct methods which make use of alterations
in frequency or phase in the received ultrasound
signals. Examples of images produced by this
latter method are the familiar pulsed Doppler B-
scan technique [3] and the new fluid velocity
vector reconstruction technique developed by
Johnson et a1 . [4].
Images produced by the first or differential
method have application in understanding moving
solid tissues such as the heart and circulatory
system and their components (e.g. , valves, ves-
sel walls, etc . ) . Any improvement in the spatial
and temporal resolution of methods for imaging
these moving structures would produce a corre-
sponding improvement in imaging the velocity of
these structures at high spatial resolution. The
effort which has been and now is being made in
the field of x-ray angiography [5] is an indica-
tion of the potential contribution which such
high spatial and temporal ultrasound imaging
methods could make to health care.
Images produced by the direct velocity imaging
techniques (based on frequency or phase shift
techniques) have application in measuring the
velocity of fluid produced or modified by tissues
(e.g . , air in the respiratory system, urine, etc. )
or the velocity of fluid tissues (e.g. , the blood).
The well-known Doppler technique (a reflection
method) measures the velocity of scattering cen-
ters (e.g. , the red and white cells) in the blood.
The new velocity vector reconstruction method (a
transmission method) measures the velocity of the
carrier (i.e., the plasma) of the scattering cen-
ters or a combination of the velocity of the two
constitutive components. Thus, the two methods
are complementary. The ability to obtain images
at high spatial and temporal resolution of blood
flows promises to be of great value for several
reasons. First, it may be argued that since the
x-ray angiographic method is of proven worth,
then any competitive ultrasound techniques would
also be valuable. Second, the new pulsed Doppler
systems now in development or use, although of
limited to moderate imaging capability, are prov-
ing to be of great value [6]. For example, the
SRI-built carotid scanner, now undergoing clinical
trials at Mayo, is proving the value of its Doppler
flow measuring features (the scanner has real-
time B-scan capability also) [7]. Third, ultra-
sound procedures are noninvasive and have no
demonstrable cumulative toxic effect when power
and energy levels are controlled. Fourth, ultra-
sound procedures are painless and usually can be
completed quickly.
236
3. Imaging Methods for Static Media
A new high resolution reflection imaging mode
has been developed in the Mayo laboratory to
image or characterize tissue structures in the
body where a full 360 degree data collection
geometry is precluded by the ribs, the spine, and
other body structures [7,8]. This new mode is
termed synthetic focus imaging [7].
The mathematics for synthetic aperture imaging
may be derived by reference to figure 1. Let
define a measure of the probability for acoustic
amplitude scattering or reflection toward the de-
tector aperture from picture element k when il-
lumination from each transmitter element in the
water tank Gmk 9jk
\
n
:\
1
\f
k
1
—
—
Tik
11
(M,M)
(M,0)
transducer array
Vijm
12 3
i(j,k,m)-l
d =
i (j,k,m) + l
Fig. 1.
Geometry for derivated of equations for
ultrasonic imaging by generalized
synthetic aperture methods. A transducer
array with N elements is in acoustic con-
tact with a water tank. Positions in the
tank are described by coordinate system
XY. A typical ray path is shown for
energy transmitted by the mth transducer
and scattered by picture element k into
the jth transducer. Here rji^ and r^k
are the distances from j to k and from
m to k, respectively. Angle 6j|< is mea-
sured from rj|<; to the Y axis counter-
clockwise. V-jjni is the voltage sample
from the jth transducer at time after
transmission of the signal by the mth
transducer. Here s and d are start and
end times of a time interval 2w which
contains the echo from k at its midpoint.
Time L is greater than the maximum time
associated with all paths defined by j,
m, and k. (Reproduced with permission
from [7].)
array aperture, measured at picture element k,
is normalized. The probability for energy scat-
tering is the square of P^. For the case of a
very narrow pulse of ultrasonic energy, P|^ is
given by
N
E
m=l
N
j = l
t+w
E
i=t-v\
^i,j,mRj,kTk,mKt-l (D
where Vi,j^ni is the ith voltage sample from the
j'^n array element when the mtn array element was
used as the transmitter. Rj^|< is the receiver-
range azimuth illumination normalization factor
between the k^n picture element and the jth array
element. Thus, Rj^|^ is independent of time, but
is an approximate function of some power of the
magnitude of rj^|< (where rj^^ is the vector from
array element j to picture e'l_ement k) and of the
cosine of the angle between fj ^ and n-; (the unit
vector normal to array element'j). Thus, R,- i,
may be written '
exp
(/a(s)ds)
- j "
q
/ds
n. • S, .
'-k
-1
(2)
where a is the attenuation coefficient along the
ray path connecting k and j, q is a rational num-
ber from 0.5 for cylindral waves to 1.0 for
spherical waves, and is the unit tanget vec-
tor at j of the ray which passes from k to j.
T)^ u, is the corresponding transmitter function
belween the k^h pixel and mth array (transmitter)
element. Thus, |^ and T|^^rn affect the energy
normalizations referred to in the definition of
P|<. N is the number of array elements. The fac-
tor K^_-j is a kernel function for performing con-
volution or cross correlation operations on the
raw data V| h If V-j is of the appropriate
pre-transmi{ted form or 'post-received processed
form, then K^.-j may be of the form of a delta
function 6q = 6t,i. The limits t ± w are
related to {he transmitter pulsewidth t^ which is
less than some arbitrary bracketing interval tB
which contains the transmitted pulse of width t^.
Then,
/ k j
ds
At
tw/2At
(3)
(4)
where At is the sampling interval time between
successive samples in the digitized signal, and u
is the effective velocity of sound in the insoni-
fied medium. The two line integrals are taken
along the rays connecting transmitter m to k and
k to receiver j respectively.
An important modification to the synthetic
focus equations above would change the sum over i
to allow for refraction or bending of acoustic
rays. The effect of this bending is manifest in
shifts in the values of s and d compared to that
obtained by computing the round trip pulse time in
a homogeneous medium. This effect is illustrated
in figure 2. The value of t would be replaced by
the correct round trip time computed using curved
rays obtained from ray tracing techniques [9,10].
The coordinate system set up by the intersection
of rays and wavefronts and the corresponsi ng
slight shift in round trip (a few percent) will
allow a low spatial resolution perturbation cor-
rection to be made to the basic homogeneous media
round trip time equations.
Another useful modification to the basic
synthetic-focusing scheme presented above would
237
Fig. 2. Orthogonal coordinate systems for describ-
ing ray paths from transducer elements in
constant (left) and variable (right) re-
fractive index media. On the left the ray
paths between each transducer element and
receiver element are segments of straight
lines. The lines of equal time of propa-
gation are circles. On the right the ray
path and equal time lines are from trans-
mitter element m to scattering site k and
the time tj|^ from k to receiver element j
is shown as a heavy line. (Reproduced
with permission from [9]).
be to replace ^■\,j,m with the output of a hard-
ware or software correlator, allowing transmission
not only of pulses of energy, but arbitrary wave-
forms as well. This may be a faster method for
obtaining the optimum resolution waveforms than
performing the inner products in software using
the kernel Kt-i as indicated by equation (1).
Figure 3 shows the similarities between the x-
ray computer assisted tomography and synthetic
focus ultrasound algorithms. It is seen that back
projection in both instances can be thought of as
a line integral in data space:
1) In the x-ray case the reconstructed value
at each pixel in the tomogram is obtained by in-
tegration along a unique path in the convolved
data space. The convolved data space is composed
of the collected projection profiles convolved
with a kernel function. The curved path associat-
ed with each pixel is a sine wave modified in
phase and amplitude for the special case of
parallel ray projections, hence, the name "sino-
gram" space for this representation of the data.
2) In the ultrasound synthetic focus method,
the image value at each pixel in the tomogram is
obtained by integration along a unique path, also
in a convolved data space. The values in the con-
volved data space can be obtained by either trans-
mitting the required waveform or by digital proc-
essing or equalization of the received data. In
the case of a homogeneous material in the object
space, e.g. , water, the set of echoes from each
receiver element for a fixed transmitter position
falls on a hyperbolic curve in data space. The
synthetic focus algorithm reconstructs the scat-
tering amplitude at each pixel by integration
along the hyperbolic curve which is unique to that
pixel .
Back projection, as just described, may be
termed "pixel driven" because the value at each
pixel is computed to completion before a value is
0 a -> TT 21T
Fig. 3. Illustration of the similarity between "back projection" reconstruction in the x-ray
case and echo mode synthetic focus image formation with ultrasound.
Left panel - Reflection Ultrasound Data Space - Pixel driven synthetic focus algorithm
generates a unique hyperbolic curve along which "back projection" algorithm performs
a line integral for the data value at each pixel value.
Right panel - Transmission X-ray Data Space - Pixel driven convolution algorithm
generates unique sine wave like curve along which "back projection" algorithm per-
forms a line integral for the data value at each pixel value.
238
assigned to a new pixel. The line integration
along curves in data space is assigned to each
pixel in this mode of computation.
An alternative conceptualization of back pro-
jection is possible in which each data point in
data space is used to influence the value of all
pixels in the image space before a new data point
is processed. This formulation may be termed
"profile driven" back projection for the reasons
just given, or termed "smear and add" back pro-
jection because the pixels which are influenced
most by each data point are located on or near a
curve in image space. The "smear" may refer to
convolution. In the x-ray case, the "smear and
add" occurs back along the original x-ray path
which produced the data value at the detector.
This is a very strong intuitive reason for the
term "back projection". In the ultrasound case,
the back projection in the image space occurs
along an ellipse (not a hyperbole) for each data
space. The ellipse, of course, represents the
locus of points of equal transmitter-receiver
pair round trip time. The superposition of the
set of values along all ellipses for all such
transducer pairs constitutes the final ultra-
sound scatter amplitude image in an analogous
manner to the formation of an x-ray convolution
algorithm image by the sum of all back projected
convolved projections. This viewpoint is further
explained in figure 4.
r-i source
Vl.5(t)
data space
14,11
(t)
-t
ul trasound
transducer array
A. Diffraction Considerations
Although no diffraction theory has been used
in the derivation of the synthetic focus equations,
it is still true that images obtained from their
application are very nearly diffraction limited
(at least qualitatively and quantitatively in
some cases). The reason for this, perhaps surpris-
ing, result lies in the use of an extremum ray
path time between the scattering center (image
point) and transducer element in the synthetic
focus method. These ray paths are often identical
to, or very nearly equal to, the characteristic
solution lines to the wave equation (the wave
equation includes diffraction effects). Thus,
although some energy travels by diffracted paths,
a large fraction usually will travel by the ex-
tremum paths used by the synthetic focus method.
B. Comparison With Seismic Methods
Many similarities exist between the synthetic
focus ultrasound algorithm developed by Johnson
et al ■ [7] and classical seismic migration methods
[ 11 ] . One similarity is in the method of collect-
ing data. Both migration and synthetic focus
methods use the concept of separate source or
transmitting points and geophone or receiving
points respectively. This is illustrated in
figure 5 for the case of a linear geophone and
circular synthetic focus ultrasound arrays.
source circle
ray source
projection
profile
convol ved
projection
profile
Fig. 4. Illustration of the ultrasound analog of back projection of a convolved
projection profile.
The left figure shows the signal voltage 5(t) received on element 5
when transmitting from element 1 as a function of time. Also shown is the
voltage V14 ii(t) for receiver 11 and transmitter 14. For simplicity, it
is assumed {hat the object space contains only one scattering point Q.
The data Vi,j(t) is assumed to be ready for back projection. The right
figure shows an x-ray source and object with a dense point at P which
produces the sharply peaked projection profile. Reconstruction of the
object is achieved by back projection of the convolved profiles for all
source positions around 360°. In the ultrasound case, the receiver echo
data is projected along ellipses in the object space (note that this is
equivalent to summing along a set of hyperbolae in the data space). As
in the x-ray case, back projection of non-negative unconvolved raw projec-
tion profile data does not give optimum resolution but instead gives the
true object convolved with a blurring function. The effect of the blurring
function may be nearly removed (deconvolved) if the projection profiles are
first convolved with a deblurring kernel. Such kernels usually have a
strong central maxima with symmetric heavily damped bipolar side lobes.
In the ultrasound case maximum resolution (the Rayleigh limit and even much
better) is achieved when the impulse response of a single scatterer is made
to look like such a kernel in the data space.
239
s
?
'f=2
.5d
transmission mode
receiver array
f J
I I
reflection mode
receiver
y [— y
' I
-1 9+1
1
= -1
DATA
COLLECTION
GEOMETRY
-3
-2 -1 0 +1 +2
—I — I — I — I —
'^^^ si^':^a'% gc gd 9e
-4
+4
^ 3
■
.d ^
0
- • c
. b
• a
DATA
DISPLAY
DOMAIN
0 12 3 4
source axis
prof i 1 e
(common source]
►common receiver
prof i 1 e
^ommon midpoint
gather
seismic section
(common offset)
DATA
COLLECTION
SCHEMES
"IL w 0 1
2 2
source angle axis —
J circular profile
I (common source)
■common receiver
circular profile
(^common midangle
gather
acoustic
circular section
Fig. 5. Illustration of the similarities between
classical seismic data collection with
linear geophone arrays and ultrasound
imaging with circular transducer arrays.
The left most column of figures illustrates
the major parameters used to describe seis-
mic data collection in one-dimensional,
i.e., linear, arrangements of source
positions (shot point = s) and receivers
(geophone = g). The right most column of
figures illustrates how the ci rcumgerence
of a circular arrangement of sources and
receivers may be described in terms of the
analogous parameters in the left column.
The shot position s is replaced by source
angle 65 and geophone position g is
replaced by the receiver angle eg.
In the top most left figure, typical source
s is located at position -3 and typical
geophone g^ is located at position +1
(arbitrary units). The mid point of the
(s,g(j) pair is at position -1 and is de-
fined as the mid point y (i.e., y = -1).
The separation of s and gj from the mid-
point is -2 and +2 respectively and is de-
fined as the offset f (f for "offset" ) .
Note f is positive when g is on the posi-
tive side of the midpoint.
Also shown is a typical seismic data col-
lection geometry with a source (dynamite
blast) located at -3 and geophones at -2,
-1, 0, 1, 2. This particular source-
receiver combination is plotted as points
a, b, c, d, e on the g-s plane. Point d
at (-3,1) in the g-s plane, represents
time history data for a source at -3 and
a geophone at +1. The midpoint and offset
coordinates (-1,2) of data point d can be
found from its projection onto the midpoint
and offset axes respectively. The direc-
tions of straight line sets of data points
at angles of 0, 45°, 90°, and 135° to the
s-axis have important properties and are
given characteristic names. A vertical
line is a prof i 1 e and corresponds to a data
point with a common source (the usual mode
of seismic collection). A horizontal line
is a common receiver profile (rarely used),
a line at 45° is a seismic section (common
offset) and is often obtained by rearrange-
ment of multiple profiles. A seismic sec-
tion is also commonly generated by towing
a geophone at a fixed separation behind a
source in the ocean. A line at 135° is
called a common midpoint gather.
The top most drawing on the right shows how
the concepts described in the left column
can be applied to circular geometries. The
240
Fig. 5. (continued)
linear parameters s, g,
angular analogues, 9$,
respectively.
and f have their
Gw, and Gf
The s-g plane, which extends to infinity
on all sides, is replaced by a finite and
bounded Qs-Qg plane (bounded by ± tt). The
Gs-eq plane provides a convenient means
for displaying the data collected for x-ray
or ultrasound transmission computed recon-
struction tomography. In the top right
figure, a circular receiving array covering
180° is shown opposite the source s. Trans-
mission data collected by this array is
contained in the trapazoids DEFG and HIJK.
Reflection data for s coincident with g is
found on the line AOB. Reflection data for
the receiver offset from the source line
AOB. Reflection data for the receiver off-
set from the source as shown is found along
lines mnp and qu. (This is one line mnpqr
when the periodic nature of the data space
is considered).
The direction of straight line data sets in
the Gs) Gg plane can be given analogous
names as those defined in the s-g plane.
These data sets are shown at the bottom of
the right column (i.e., circular profile,
common receiver circular profile, common
midangle gather, and acoustic circular
sections) .
typical
signal
reflection, scatter and
Doppler Scatter space
above EAT surface
earliest arrival
time (EAT) surface
transmi ssion
mode array
domain null space
beneath EAT surface
Fig. 6. Illustration of the mathematical proper-
ties of acoustic data collected from
source-receiver pairs on a circular
boundary. The time dependent voltage
of the signal received by receiver g is
plotted as an amplitude (fourth amplitude
dimension not shown) v_s_. , Gs, Gg, and
time t. The coordinates 9$, Gg are
source and receiver positions as defined
by the drawing at the right. For any two
points s and g on the circumference, there
exists an earliest arrival time (EAT) for
acoustic energy to propagate from s to g.
This time is zero if s is coincident with
g but is maximum (or near maximum for non-
homogenous substances) when s and g are on
opposite ends of a diameter. Thus, for
times t 1 ess than EAT, the values in the
range (not the domain) of (Gs, Gg, t) are
The similarity between linear array seismic
methods and synthetic focus medical ultrasound
methods exists for both linear and circular
scanning geometries. Circular geometries are
found in x-ray computed tomographic reconstruc-
tion instruments and in their ultrasound counter-
parts developed at Mayo Clinic by Greenleaf and
Johnson [121. This similarity is explored and
developed in figure 5 and its legend. Many of
the important features of transmission mode
reconstruction imaging such as "sinogram" space,
or transformation from fan beam to parallel beam
geometries, can be analyzed with the aid of the
well-known seismic s-g plane data representation
(s-g means source and geophone pair location for
each data record).
zero (no signal has as yet arrived). Thus,
the EAT value assigned to each pair (Gs,
Gg) defines a surface below which all
values in the same range are zero. This
surface is called the EAT surface. The
region above the 0-z axis (2j_e. , the
9s ~ 9q line) provides maximum time separa-
tion of echoes from most neighboring points
and provides maximum Doppler shifts for
fluid flows. The region above areas abed
and ehij corresponds to receiver positions
nearly opposite from the source and are
used for computed tomographic (i.e., C.T.)
transmission data collection. A maximum
arrival time (MAT) surface (not shown)
also can be drawn for which all points
above this surface correspond to multiple
reflections or multiple scattering events,
or both.
The nature of the s-g data domain representa-
tion in circular geometries can be extended to
include the time variable. An explanation of the
features of this three-dimensional data domain is
given in figure 6. There it is shown that the
domain can be partitioned into two regions by a
surface which corresponds to the earliest arrival
time of an acoustic pulse or signal. The distance
between a point P on this surface and the s-g
plane corresponds (nearly inversely proportional-
ly) to the "merit" of the echo spatial resolution
information contained in the signal associated
with the point P (i.e., a receiver opposite a
source can have no or little echo information).
The synthetic focus method reported by Johnson
et al . [7], according to the seismic nomenclature.
241
collects a profile for each transmitter position.
The side looking radars or side looking synthetic
aperture radars can be classified in seismic terms
as producing one zero-offset section per flight
path. It is well-known that superior focused
(i.e., "migrated") seismic images can be obtained
by the use of multiple profiles rather than by the
use of one zero-offset section [11]. Thus, our
synthetic focus method should produce images
superior to those produced by literal adaptation
of side looking radar or synthetic aperture tech-
niques to medical imaging.
Also, in radar technology, refraction (i.e.,
normal moveout corrections) are not usually, if
ever, made (but should be for medical imaging).
4. Imaging in Moving Media
A. General Considerations
Imaging in the presence of moving media re-
quires that certain changes be made to the basic
image forming equations for both reflection and
transmission modes of imaging. We will show that
consideration of these requirements provides not
only a method for imaging structures embedded in
the moving media (points in the moving media) but
also methods for imaging the velocity distribu-
tion of the moving media itself.
The equations of synthetic focusing can be
adapted to the case of moving media by using the
time of travel associated with the ray paths
which have been bent by the moving media. Thus,
the expression for the time t of travel must be
replaced by an expression containing the velocity
of the media. For the case where the velocity of
the media is small compared to the acoustic speed,
this may be written as (by modifying equation 3)
m k
Here, u is acoustic speed, \f is fluid velocity,
and T is a unit tanget vector along the ray path
connecting k to the transducer elements m or j. A
more exact expression for t, in the case when t
is not very small in comparison to u, has been
given by Johnson and others [10,13]. In addition
to time shifts, a moving media also produces a com-
pression or expansion of the shape of a waveform.
When more exact imaging is required, the effect of
this stretching may be removed by use of an appro-
priate compensating algorithm as is now shown.
Let g(y) = b • h(t-to) be the received synthetic
focus waveform for \? = 0. Here, b is the scat-
tering strength and h is the transmitted waveform
(for simplicity, the frequency dependence of the
scattering process is assumed to be found in
H (•))• Let an upper case letter represent the
Fourier transform of a corresponding lower case
letter. Then, the corresponding expression in
frequency space is G(f) = b • exp(-2iTrfto)H(f ) .
The Doppler frequency shift due to motion of
the media has the effect of replacing f by (f + Af).
I^lere, Af is given by Af = f ^ • (Tr + Ts)/u, where
V is the velocity of the flow and where Tr and Ts
are the unit tanget vectors from the scattering
center in the flow to the receiver and transmitter
(source) respectively. For |V|/u <<1, the effect
of this frequency shift changes the form of G(f)
to GD(f) given by GD(f) = b • exp(-2^fto)H(l - 6)(f)
where 6f = Af. The corresponding time functions
are given by
gp(t) = (b/|l - 6|)h((t - to)/(l - 6)). (6)
Thus, movement of a scattering center toward a
source and receiver produces a shortening of pulse
width and an increase in pulse amplitude.
B. Synthetic Focus Doppler Imaging
The principle of synthetic focusing may be ap-
plied to Doppler velocity imaging to improve the
spatial and velocity resolution and to capture
more of the Doppler scattered signal (by the use
of a larger receiving aperture) and thereby in-
creasing the signal-to-noise ratio in the final
velocity image. Simply increasing the aperture
of a single transducer may not improve Doppler
resolution and signal-to-noise ratio because the
Doppler shift will not be constant over the large
aperture. A method is needed which will make
use of the different Doppler shifts on a large
sampled aperture.
For simplicity, a method suitable for flows re-
stricted to a plane will be given first. It is
assumed that the transmitting and receiving trans-
ducers are also restricted to this plane and are
located on part of, or on the complete, circum-
ference of a circle. Assume that the transmitter
produces a narrow beam of pulsed-energy which
constitutes a cord of this circle. Let the velo-
city of the flow at e|ch point along the cord be
^(q). Let Ts(q) and Tj(q) be the unit tanget
vectors from the scattering center q to the trans-
mitter source s and receiver j respectively. The
use of time gates to achieve spatial resolution is
assumed. Then the Doppler shift Af is given by
(f/u)V • (Ts + Tj). This may be written as
Af(q) = (f/u(q)) [(cos ot + cos ej)Vx + (sin et +
sin ej)Vy]q. Then this constitutes a set of
simultaneous equations in Vx and Vy. This set is
usually over determined since the set can be
spanned with only two independent values of e j ,
but the over-determination provides for better
signal-to-noise level. The above equation may
also be written as
Af(q)j = [f/u(q)] (|V(q)|) [cos(\r,fs) (7)
+ COs(V,Tj)], j = 1,2...
Note that the first term is a frequency bias
and the second term has a period of 2-n in ej.
The direction of the flow ? and its magnitude may
be found from this set of equation by several
schemes :
1) The Fourier transform of the |.bQve equation
may be taken with respect to angle (V,Tj). The
square root of the sum of the squares of sin and
cos terms with pet;ii}d 2-n is equal to (|V|f/u)
while the angle (V,Ts) is proportional to the arc
tangent of the ratio of these terms.
2) A least squares fit of a phase shifted
cosine function can be made to the data. This
least squares technique can also be applied to
three-dimensional flows and two-dimensional de-
tectors. The velocity V can then be obtained.
These two schemes and the theory presented up
to this point assume that a unique value of Af
may be obtained for each point q and detector j.
242
This is not exactly true because in practice, the
frequency measure is broadened due to: 1) noise,
2) ambiguity due to range gating, and 3) turbu-
lence in the flow. In most or many cases, the
broadening may be tolerated by the use of offset
frequency quadrature Doppler detection schemes
which produce an output voltage whose mean is
proportional to the mean Doppler shift of a
broadened signal [3].
C. Flow Reconstruction by Acoustic Transmission
We have previously suggested that fluid flow
within a measurement region may be determined by
transmitting and receiving acoustic energy through
the measurement region along a plurality of rays
such that each volume element is transversed by a
set of rays having components in each direction
for which flow components are to be reconstructed
[10]. The propagation time of the acoustic energy
along the plurality of rays constitutes the only
measurements required by this method which has now
been verified for flows with velocities small com-
pared to the speed of sound [4]. Since this
method is analogous to computed reconstruction
tomography methods and has been reported elsewhere,
the corresponding theory is not reported here.
D. Synergistic Flow Reconstruction
It is clear that both transmission reconstruc-
tion and Doppler methods could be applied simul-
taneously with sufficient parallel data collection
and processing capability. Such an approach, com-
bining larger apertures and both imaging modes,
although perhaps impractical at this time, has
several theoretical advantages: 1) greater signal-
to-noise ratios, 2) greater spatial and velocity
resolutions, 3) the transmission reconstruction
mode can provide refractive index information for
correction of the Doppler imaging mode, and 4)
drag velocity might be calculated from the dif-
ference between the Doppler mode velocity image
(scattering center velocity) transmission mode
velocity image. These principles could be applied
to statistically steady-state flow using one
Doppler channel and one transmission channel by
time multiplex methods.
5. Computer Simulation and
Experimental Studies
An example of the resolution capability of the
synthetic focus technique is demonstrated in
[ figure 7. This figure is very significant because
I it presents experimental evidence that resolution
1 of less than one-half wavelength (at the center
I frequency of a pulse) can be obtained with ultra-
1 sound; at 3 MHz this corresponds to 0.25 rrm. It
I is expected that, when synthetic focusing, refrac-
I tive index reconstruction and other ultrasound
imaging modalities are combined, a synergistic
union will result. Thus, each mode will not only
add its own form of new information but will also
help remove some of the limitations of other
' forms of imaging. For example, knowledge of re-
) fractive index can correct for defocusing effects
i in echo imaging mode [9,11].
Thus, a major limitation to applying the com-
1 plete capabilities of the synthetic focus tech-
nique to complex tissues can be solved by the
synergistic treatment of the problem of refrac-
raw data deconvolved data
Fig. 7 Synthetic focus images of a nylon thread
from data taken from one view angle and
from multiple view angles. The top left
image represents the raw data from 26 re-
ceiver elements for each of five trans-
mitters. Left image A is the correspond-
ing synthesized image. The lower left
image B is obtained from simulating the
effect of many such 32 element arrays on
a circle or radius corresponding to geo-
metry of image A. Top right is the re-
sult of deconvolving the ray data of top
left with impulse response of the system.
Middle right A and lower right are images
synthesized from deconvolved data and cor-
responding to A and B, respectively. The
labeled line segments and circles in the
center are the relative wavelength of
sound drawn to the same scale as the re-
constructed images. Note that the full
width at half maximum of the peak in image
B' (FWHMg') is less than one-half wave-
length (a/2) while that in image B is
larger and more nearly A/2. Each picture
element is .04 mm square. The resonant
frequency of the transducer elements is
about 3 MHz. (Reproduced with permission
from Johnson et al . , Digital Processing
of Biomedical Images, 1976, pp. 203-226.)
tion of ultrasound. Consequently, an iterative
correction of refraction has been developed. This
ray tracing technique has also been approached
from established geophysical seismic imaging
theory. The corrections used by seismologists to
unscramble their seismographic results indicate
the feasibility of this approach. An example of
the order of improvement attending images origi-
nally collected by synthetic focus methods, when
seismic image correction techniques are used, is
given in figure 8. This image is significant be-
cause it represents (to our knowledge) the first
application of seismic wave migration techniques
to real ultrasound data. This image has been
presented in an earlier paper [9].
In these images presented in figures 7 and 8,
no attempt has been made to accurately control
the waveform of the echo from a single point to
maximize resolution. Although, the right-hand
243
plastic wedge part of "MAYO"
Fig. 8 Refraction corrected synthetic focus image.
The bottom portion of figure shows a
plastic wedge and pattern of nylon threads
which spell "Mayo", however, only the "MA"
portion is visible in the figure. An ar-
ray of 32 transducer elements is located
to the left but is not shown. Sonic waves
produced by the array refract through the
wedge, reflect or scatter from the threads,
pass back through the wedge and are then
received by the array. No recognizable
image was produced when the corresponding
data was processed with a simple synthetic
focus algorithm which assumed constant
speed of the ultrasound waves. The out-of-
focus image produced by this simple algo-
rithm is not shown here.
The image at the top is the result of ap-
plication of a geophysical data processing
algorithm which corrects for refraction
without prior information of the presence
of the wedge. This is accomplished by
"wave front migration" techniques which
determine the effective refractive index
layer by layer as the algorithm works it-
self away from the array. Such algorithms
work for layered materials but could be ex-
tended to structures found in the human
body. The center frequency of the damped
burst transmitted from the array is 3 MHz.
The target is the same one reported in
Volume 6 of Acoustical Holography by S. A.
Johnson et aT. [7 ] .
side of figure 7 shows the image produced by a
deconvolution process which satisfies only the
general requirements. Therefore, a computer
simulation study was undertaken to investigate
the maximum 2-D resolution limit which can be
obtained with a signal constrained to have no
frequency components greater than an upper fre-
quency limit fm- This transmitted kernel signal
h(t) was chosen to have a frequency distribution
given by H(f) = Hglfl for |f| < fm and H(f) =
0 for |f| > fm. This function is the well-known
Ramachandran Lakshminaraynan kernel [iSl. A data
collection geometry of a 12 cm radius ring of 120
common transmitter and receiver positions equally
spaced on 360 degrees was simulated. Imaging
was confined within a concentric 7 cm circle. The
value of fjTi was set at 1 MHz and the simulated re-
ceived echo signals were sampled at 100 ns inter-
vals. The simulated received echo signals were
back projected along ellipses in the image space
as per figure 4. The 2-D point response function
(2-D PRF) was found not to differ significantly
between points near the center and edge of the
7 cm radius imaging region. Various (time) fre-
quency filtered versions of the above kernel were
tried also. Only slight improvement in 2-D PRF
was observed with some kernels but reduction in
2-D side lobe response could be obtained with an
attendant slight loss in resolution with other
kernels. Some of the results of these studies
are presented in figure 9.
Since the experimental image obtained by ap-
plication of the new transmission mode fluid
temperature and vector velocity technique have
been published in more detail previously [4], only
one example reconstruction is presented here.
Figure 10 shows reconstructions of fluid flows
made in the refractive index reconstruction breast
scanner at Mayo Clinic. Applications of these
techniques, for example, for measurement and re-
construction of tissue temperatures during hyper-
thermia cancer therapy have been reported else-
where [16].
Data on X.i£.2 Image on
Fig. 9 Illustration of simulated data and cor-
responding synthetic focus resolution test.
Top left: gray scale plot of received
data for two scattering centers separated
by 1/2 wavelength (measured at the maxi-
mum frequency limit of transmitted kernel).
The amplitude for 120 views on 360° vs
time is shown. Bottom left: amplitude
plot of above data along brightened line
l]^l2. Top right - 0 mm by 9.6 mm synthetic
focus image (64 by 64 pixels) from these
data. Bottom right: amplitude plot along
the white line I3I4 in the above image
demonstrating the resolution of the two
point targets. In the gray scale images
above, gray = zero, black = negative, and
white = positive amplitudes.
244
Fig. 10. Experimental data and reconstruction of vector components of fluid
vortex. Top left shows experimental data and is an image of the
difference between the time of arrival with flow and without flow
(fast arrival = white, no change gray, slow = black) vs^. scan
position (left to right) vs_. angle of view (top to bottom). Top
right shows reconstructed planar fluid speed (v^ + V^)''5, black is
zero, white is positive. Bottom left shows x component of velocity
Vx- Bottom right shows y component of velocity Vy. In Vx and Vy,
black is negative, gray is zero and white is positive. Reconstruced
flow is maximum (73 cm/s) at a radium of 0.62 cm. Reconstructions
are 64 pixels per side. Geometry of vortex scanning plane, and
vortex generator are shown in top and side views in right margin.
(Reproduced with permission from Greenleaf et al . , Quantitative
Imaging from Transmission Ultrasound (in press ) . )
6. Conclusion
We suggest that instruments capable of produc-
ing higher resolution images of specific ultra-
sound tissue parameters would provide a valuable
service in clinical applications. Wide aperture
techniques are known which can provide this high
resolution. To be successfully applied in the
human body, these wide aperture techniques must
be modified to compensate for refraction and pos-
sibly, in some cases, also diffraction effects.
Candidate techniques can be found in the synthetic
focusing methods developed in our laboratory and
in analogous geophysical or seismic methods.
These techniques can provide resolution of one-
half wavelength in homogeneous, isotropic media.
It seems reasonable that this degree of resolu-
tion may De approached in the body by use of wide
apertures and improved refraction and absorption
compensating algorithms. Improvement of resolu-
tion from one wavelength to one-half wavelength
is difficult because knowledge of the spatial and
temporal history of the transmitted and received
waveform is necessary for application of these
algorithms. This knowledge may, in principle,
be obtained with the proper transducer design and
scanner geometry. Our computer simulation and
laboratory experiments provide encouragement and
evidence that resolutions of about one to one-
half wavelength should be feasible in some ap-
plications. We further predict that efforts to
design economical scanners to provide this resolu-
tion will be a fruitful research and development
endeavor. We believe such endeavors will rely
heavily upon the use of high speed digital data
processing methods and sophisticated mechanically
and electronically scanned apertures with large
solid angles.
Acknowl edgment
The support from Dr. Earl H. Wood, Dr. Erik
Ritman, and the staff of the Biodynamics Research
Unit, Mayo Clinic, is appreciated. The refraction
corrected image of the "MA" shown in figure 9 was
computed using seismic migration techniques with
the help of John Parr, of Houston, Texas, and his
help is greatly appreciated. This research was
supported by Grants HL-00170, HL-00060, RR-000Q7.
Hl-04664, NIH-HT-42904 from the National Insti-
tutes of Health, United States Public Health
Service; NCI-CB-64041 from the National Cancer
Institute.
References
[1] Wolfe, J. N. , Risk for breast cancer develop-
ment determined by mammographic parenchymal
pattern. Cancer 37. 2486-2492 (1976).
[2] Wolfe, J. N., Breast patterns as an index of
risk for developing breast cancer. Am. J.
Roentgenol. 126, 1130-1139 (1976).
245
[3] Woodcock, J. P., Development of the ultra- [lO]
sonic flowmeter. Ultrasound in Medicine and
Biology 2. 11-18 (1975).
[4] Johnson, S. A., Greenleaf, J. F. , Tanaka, M.,
and Flandro, G., Reconstructing three-
dimensional temperature and fluid velocity
vector fields from acoustic transmission [ll]
measurements. Instrument Society of America
Transactions, March 1977 (in press).
[5] Sandler, H. and Rasmussen, D. , Angiographic
analysis of heart geometry, in Roentgen- , [i2]
Cine-, and Videodensitometry, Paul H.
Heintzen, ed., pp. 212-223 (Georg Thieme
Verlag, Stuttgart, 1971).
[6] Cardiovascular applications of ultrasound.
Proceedings of an International Symposium
held at Janssen Pharmaceutics, Beerse,
Belgium, May 29-30, 1973. Robert S. Reneman,
ed. (American Elsevier Publishing Company,
Inc., New York, 1974). [13]
[7] Johnson, S. A., Greenleaf, J. P., Duck,
F. A., Chu, A., Samayoa, W. F., and Gilbert,
B. K., Digital Computer Simulation Study of [14]
a Real-Time Collection, Postprocessing
Synthetic Focusing Ultrasound Cardiac Camera,
in Acoustical Holography, Newell Booth, ed..
Vol. 6, pp. 193-211 (Plenum Press, New York,
1975).
[15]
[8] Duck, F., Johnson, S. A., Greenleaf, J. F. ,
and Samayoa, W. , Digital image focusing in
the near field of a sampled acoustic aper-
ture. Ultrasonics, 83-88 (March, 1977).
[9] Johnson, S. A., Greenleaf, J. F., Bahn,
R. C, Samayoa, W. F., Duck, F., Hansen, C, [16]
Tanaka, M., and Lent, A., Reconstruction of
three-dimensional ultrasound tissue proper-
ties from transmission and other data.
American College of Radiology Conferesce on
Computerized Tomography and Radiology, St.
Louis, Missouri, April 25-28, 1976, pp.
149-162.
Johnson, S. A., Greenleaf, J. F., Samayoa,
W. F., Duck, F. A., and Sjostrand, J.,
Reconstruction of three-dimensional velocity
fields and other parameters by acoustic ray
tracing. 1975 Ultrasonics Symposium Proceed-
ings, IEEE Cat. #75 CHO 994-4SU.
Claerbout, J. F., Fundamentals of Geophysi-
cal Data Processing with Applications to
Petroleum Prospecting (McGraw-Hill, New
York, 1976).
Greenleaf, J. F., and Johnson, S. A.,
Algebraic Reconstruction of Spatial Distri-
butions of Refractive Index and Attenuation
in Tissues from Time-of-Fl ight and Amplitude
Profiles, in Ultrasonic Tissue Characteriza-
tion, M. Linzer, ed., National Bureau of
Standards Spec. Publ. 453, pp. 109-119,
October, 1976. (U.S. Government Printing
Office, Washington, D.C., 1976).
White, R. W., Acoustic ray tracing in moving
inhomogeneous fluids. Acoustical Society of
America 53 (6), 1700-1703 (1973).
Brooks, Rodney A. and Di Chiro, Giovanni,
Principles of computer assisted tomography
(CAT) in radiographic and radioisotopic
imaging, Phys. Med. Biol. 2j[ (5), 689-732
(1976).
Ramachandran , G. N., and Lakshminarayanan ,
A. v.. Three-dimensional reconstruction from
radiographs and electron micrographs: ap-
plication of convolutions instead of Fourier
transforms, Proc. Natl. Acad. Sci., U.S. 68
(9), 2236-2240 (1971).
Johnson, S. A., Greenleaf, J. F., Rajago-
palan, B., Christensen, D. A., and Baxter,
B. , Nonintrusive acoustic temperature tomo-
graphy for measurement of microwave and
ultrasound-induced hyperthermia. Workshop
on Electromagnetics and Cancer, 1977 lEEE-
MTTS International Microwave Symposium,
June 24, 1977. (Submitted to the Journal
of Bioengineering . )
246
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed., National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
MAPPING TRUE ULTRASONIC BACKSCATTER AND ATTENUATION DISTRIBUTION IN TISSUE -
A DIGITAL RECONSTRUCTION APPROACH
F. A. Duck and C. R. Hill
Physics Divisiorij Institute of Cancer Research
Clifton Avenue, Sutton, Surrey, United Kingdom
The back-scattered intensity from any location in an ultrasonically irradiated
medium is dependent upon both the local back-scattering cross-section and the attenua-
tion by the overlying medium. An iterative digital reconstruction technique has been
investigated which is capable of mapping quantitatively and separately the distribu-
tions of attenuation coefficient and back-scatter cross-section using multiple pulse-
echo data. Such a technique will enable tissue to be characterised on the basis of
these two separate acoustic parameters. A description of the technique is given,
together with results from simulation studies leading to an improved processing tech-
nique. The potential and limitations of the method are discussed.
Key words: Attenuation; back-scattering cross-section; digital reconstruction;
iterative; ultrasound.
1. Introduction
A major deficiency of currently used pulse-echo
techniques for tissue characterisation and imaging
is their failure to give truly quantitative mea-
surements of tissue-specific acoustic parameters.
The problem arises since the observed time-
dependent echo amplitudes result from two sets of
unknown quantities. These are back-scattering
cross-sections of the interrogated volumes and
attenuation coefficients of the volume in the
transmission path. In conventional B-Scan prac-
tice it is usual to emphasise the use of back-
scattering cross-section as the useful tissue
characterisation parameter, and suppress, or com-
pensate for, the amplitude changes due to the tis-
sue attenuation. This latter is done in two ways.
In the first place, a time-gain-compensation (TGC)
amplifier is used which compensates for attenuation
under the assumption that attenuation, if not uni-
form, varies in a way which is identical irrespec-
tive of transducer position and orientation. The
precise form of the attenuation versus time (or
depth) variation is often under operator control.
Secondly, compound scanning techniques are used
which result in an evening-out of the shadowing
effects of overlying tissue layers, with any given
TGC function. Automatic methods have been de-
veloped for adaptive gain compensation (McDicken
et al. [l]i; De Clercq et al. [2]). These have
shown that some improvement in gain compensation
can be obtained, and presumably the TGC function
generated could be used diagnostical ly. However
such techniques are designed to operate on multiple
^Figures in brackets indicate literature
references at the end of this paper.
pulse-echo data taken from a particular orientation,
assume uniform back-scattering and have an operat-
ing time constant of several seconds. In contrast,
the method we have developed, and which is describ-
ed below, generates separately two-dimensional
maps of attenuation coefficient and back-scatter-
ing cross-section, from multiple orientated pulse-
echo data which may be gathered very rapidly for
subsequent processing. The two clear incentives
for investigating this technique are:
1) It provides a quantitative map of acoustic
attenuation coefficients which may be used dfag-
nostically for tissue characterisation.
2) It provides a correctly attenuation-
compensated B-Scan, a map which can now be correct-
ly referred to as a back-scatter map.
2. The Method
As noted above little or no diagnostic informa-
tion from local attenuation variations is avail-
able when compound B-Scanning is used. In simple
scanning, however, regions which differ signifi-
cantly from the expected local attenuation will,
by virtue of the pre-set TGC used throughout the
scan, result in modified signal levels from re-
gions beyond them. For example a local, dense
athero-sclerotic lesion will cast a shadow below
it, and a fluid filled cyst of low attenuation will
result in increased amplitude signals from tissue
beyond it since the gain is over-compensating for
attenuation losses. Such signs are used clinically
for diagnosis but only under those circumstances
where the local effects are clear. The basic con-
cepts are shown simply in diagrammatic form in
figure 1, which shows images of a localised in-
crease in attenuation (at A) in a uniform scatter-
ing medium. A simple linear scan results in an
247
Primary images
a) Linear scan 1 b) Linear scan 2
Secondary images
c) Scatter map d) Attenuation map
vestigated in ultrasound for the reconstruction of
acoustic attenuation from transmitted pulse ampli-
tudes (Greenleaf et al . [3]) and for the reconstruc-
tion of acoustic refractive indices from transmit-
ted pulse times-of-fl ight (Greenleaf et al . [4]).
The analytic description for this problem was
formulated by Radon [5] and a range of computation-
al techniques, based on this analysis or on the
use of iterative techniques, have been reported
(e.g. , Gordon et al . [6]). In transmission imaging
the reduction in intensity (or shadow) in the
transmitted energy can be measured directly. In
the present situation however this reduced intensi-
ty signal cannot be measured directly but is re-
radiated with a strength depending upon the local
back-scatter cross-section. A new set of unknown
quantities is therefore introduced. Although the
values of these back-scattered cross-sections are
unknown their related positions in space are known
since the position and orientation of the trans-
ducer and time delay can be measured.
3. Theory
Fig. 1. The generation of separate maps of back-
scatter cross-section and attenuation co-
efficient from multiple pulse echo data.
A local increase in attenuation at A with-
in a scattering medium casts shadows S and
T in simple linear scan images (a) and (b).
The combination of these by peak detection
gives a first guess at a scatter map (c).
The differences as between this and a, b
are used to generate d, the attenuation
coefficient map and correct c.
image (fig. la) in which a shadow (S) is cast
through the image behind the local attenuating
site. In this simple case it would normally be
assumed that there was indeed a locally high at-
tenuation at A. However, an identical image could
have been generated if there was an elongated re-
gion of low back-scatter cross-section which was
coincident with the shadow region S. These two
situations can be easily distinguished by carry-
ing out a second simple linear scan at a new angle
(9) (fig. lb). Now the region which was in shadow
(S) is imaged clearly and there is a new shadow
(T). Combining the two simple scans gives a com-
pounds scan (fig. Ic) which may destroy the at-
tenuation information. Comparing the simple and
compound scans enables the true attenuation map to
be inferred.
In general of course, all tissue layers will
cast an acoustic shadow on tissue layers behind
them to a greater or lesser extent, and the real
situation is additionally complicated because the
scattering characteristics of the medium vary.
However, as seen below, account can be taken of
both sets of parameters simultaneously.
The physical situation described is in many
ways conceptually similar to the problem handled
in axial tomography, and whose computed solutions
are exploited widely in clinical medicine for the
reconstruction of x-ray attenuation coefficients
from transmitted x-ray intensities. Here the
problem of the reconstruction in 3 dimensions of
a parameter map from a set of 2-dimensional pro-
files is reduced to that of the reconstruction of
a set of parallel, 2-dimensional distributions of
the parameter from related 1-dimensional profiles.
Such transmission imaging techniques have been in-
The mathematical formulation of the problem is
designed to establish a set of simultaneous equa-
tions based upon an assumed model; the equations
may then be solved iteratively by computational
techniques. At present there is no equivalent
formulation to that of Radon for the reconstruc-
tion of distributions of parameters from transmit-
ted profiles. Indeed it may well be that if there
were it would not be of direct use computationally
because of the necessary assumptions made in the
model .
The back-scattered intensity Ip received from
any position r in an ultrasonically irradiated
field, can be given as
I = I
ref
A(r)
exp
a It'
r
(1)
where A(r) is a factor depending upon the beam
geometry;
r = the distance of the scattering target from
the acoustic centre of the transducer;
c = the local attenuation coefficient; and
a = the local back-scattering cross-section per
unit volume per steradian.
In logarithmic forms this becomes
2.n
(2)
In order that the situation is amenable to a digi-
tal solution the region to be imaged can be model-
led as a rectangular grid of n elements, within
each of which and a are constant, with the only
changes being at the boundaries of the elements.
This geometry is illustrated in figure 2. In ad-
dition it is assumed that there are no losses from
absorption and scattering processes outside the
region to be imaged. Then, for any image element
within this set, eq. (2) can be rewritten as
In
(Vef)
= K
2L + ^n(0^
n=l
(3)
248
—
— i
i
Transducer
Fig. 2. Image grid geometry. The image consists
of n cells Pi...Pn. Each cell Pf has as-
sociated with it acoustic parameters of
attenuation coefficient a-j and back-
scatter cross-section, o-jWp^ is the ray/
cell intersection length.
where K,- is a potentially known geometric factor,
and where W-j , i = l...n, are a set of geometric
weighting factors related to the image grid and
beam geometry. If ray geometry is assumed then
W-j is the length of intersection of the ray with
the image element Pi. In general W-j = 0 for
many i.
If a multiplicity of measurements 1^ are made,
from a number of directions then eq. (3) extends
to a set of simultaneous equations in which a-j,
i = l...n and a-j, i = l...n are the unknowns.
One characteristic of this set of equations is
that it is sparse since, as pointed out above,
many values of Wi are zero. A digital iterative
technique is suited to the solution since it can
handle such large sets of equations under condi-
tions where exact solutions may be multiple, or
not exist at all. As pointed out above, such
techniques have already been used highly success-
fully in medicine for the reconstruction of the
distribution of x-ray attenuation coefficients
from transmitted profiles and in radiation emis-
sion tomography.
4. Computational Procedure and Results
As presented, the problem is closer to that
posed in emission than in transmission tomography.
In the former case the isotopic distribution is
to be mapped in the presence of attenuation. In
general however the techniques described in emis-
sion tomography either ignore the effect of at-
tenuation or compensate for it simply by comparing
opposed views. In general no serious attempt is
made to generate attenuation maps for diagnostic
use. Clearly, our formulation set out above al-
lows for the computation of both back-scatter
cross-sections and attenuation coefficients
simultaneously. The successful implementation of
this approach is described later.
A variety of iterative approaches are available
and are under investigation. The technique which
at present appears most promising and has been in-
vestigated most fully generates the (n + l)th
value of attenuation coefficient from the nth
value a" by
n+1
W
1 +
Z(W2) +
V)
(4)
where e is the difference between the data value
from any location and the predicted value from the
current image arrays, and c is a weighting factor
for the back-scatter cross-section. (WH is sum-
med only over the image elements in the beam up to
the position related to the data value used. Simi-
larly, the (n + l)th value of o.j is generated from
the nth value using eq. (5)
n+1
(5)
These expressions are similar to those used by
Gordon et al. [7] in the transmission reconstruc-
tion algorithms ART.
The program set up to investigate this approach
can receive either simulated or real data. Simu-
lation enables up to 10 circular fields of speci-
fied radius, position and acoustic properties to
be used. Figure 3 shows a reconstructed image
using simulated data from a cylindrical scattering
region containing smaller cylindrical regions of
increased attenuation. The larger region is of
12 mm radius and has an attenuation coefficient of
Fig. 3. Reconstructed attenuation coefficient map
from simulated data after 4 cycles. The
image is on a 33 x 33 matrix of 1 mm square
cells. The object was a cylindrical region
12 mm diameter, attenuation coefficient
0.2 neper cm"i including 4 regions 4, 3,
2, 1 mm diameter, attenuation coefficient
0.1 neper cm"^. Data was from 10 linear
scans with pulse spacing 0.5 mm. The as-
sociated scatter field was uniform.
249
0.2 neper cm"i. The smaller regions have radii
4, 3, 2, and 1 mm and attenuation coefficients of
0.1 neper cm"i. The image shown was generated
after 4 iterative cycles, i.e. after all the data
had been used 4 times. The convergence and numeri-
cal accuracy of the computation is indicated in
figure 4. The attenuation coefficient values in
pixels along column 18 are plotted for the second,
fourth and sixth iterative cycles. There is little
alteration in the values from the third to the
fifth cycle and the solution clearly is converging
to values which are numerically correct.
Such rapid convergence to a low noise image is
dependent very largely upon the choice of iterative
procedure used. In particular it has been found
that the most rapid convergence has been from a
primary scatter map generated from all the data
either averaged or peak detected, with a primary
attenuation map set to zero. Convergence did oc-
cur from a primary scatter map generated from a
single linear scan, but was very slow and oscilla-
tory in pattern. The sequence of iterative cor-
rections has also been observed to alter the noise
characteristics of the final image. Several se-
quence orders have been investigated including se-
quencing along a pulse train, in a forward or re-
25
30
5 10) 15 20
Image cell
Fig. 4. The convergence of attenuation values along
column 18 of figure 3 after 2, 4 and 6
cycles compared with the simulated object
values. The starting assumption was a = 0
everywhere. Pulses taken in sequence and
pulse values in reverse sequence.
z!5
Fig. 5. Reconstructed maps of backscatter cross-section (upper) and attenuation coefficient (lower)
from overlapping regions. The objects are shown on the left. The reconstructed maps with
high threshold are shown in the center. An edge artifact appears in the attenuation map.
Low threshold images (right) are less noisy and show suppression of the artifact.
250
verse direction, and changing order in which pulses
are used. An additional weighting not included in
eqs. (4) and (5) has also been used which improved
convergence, to compensate for the fact that pixel
values close to the transducer are modified more
often than those at a distance along a sampled
pulse.
Abrupt changes in the scatter level have been
found to result in edge artifacts occurring in the
attenuating field. This is illustrated in figure
5, which shows reconstruction of a 3-component sim-
ulated field with overlapping regions of altered a
and a. A ring artifact in the attenuation plot is
clearly evident, associated with the perimeter of
the small scattering region. However, when a
threshold is introduced to limit the conditions
under which eqs. (4) and (5) operate (as in fig. 5,
right) the edge artifact is suppressed. It is
clear that in tissue it is unlikely that such an
abrupt change will occur but such a threshold also
enables a control to be kept on the effect of high-
ly directional spectral reflections.
5. Discussion
A convergent iterative computational procedure
has been developed to solve the set of equations
describing quantitatively the back-scatter of
sound from a distributed scattering and attenuat-
ing medium. It has been shown that, under simula-
tion conditions, accurate reconstructed maps of
both back-scatter and attenuation distributions
can be obtained. Some problems posed by the ap-
plication of the techniques to real data have al-
ready been investigated. Since an iterative pro-
cedure is being used, it is possible to include
controls which limit the range of application with-
in the iterative process to include only those
which are deemed to force the process to converge
effectively. It is clear however that the success
of the application in tissue will depend upon the
extent of validity of some of the assumptions used
in the model, and this is yet to be evaluated.
Anisotropy in o or a, position errors caused by
refractions and the effects of a finite beam width
and diffraction errors must all place some ulti-
mate limitation on the method. Within these limi-
tations however the techniques offer the potential
of a quantitative imaging of back-scatter cross-
section and attenuation coefficients from pulse-
echo data.
6. References
[1] McDicken, W. N., Evans, D. H., and Robertson,
D. A. R. , Automatic sensitivity control in
diagnostic ultrasonics. Ultrasonics 12, 173-
176 (1974).
[2] DeClercq, A. and Maginness, M. G. Adaptive
Gain Control for Dynamic Ultrasound Imaging,
in IEEE Ultrasonics Symposium Proceedings
pp. 59-63 (1975).
[3] Greenleaf, J. F., Johnson, S. A., Lee, S. L.,
Herman, G. T., and Wood, E. H., Algebraic
Reconstruction of Spatial Distributions of
Acoustic Absorption within Tissue from Their
Two-Dimensional Acoustic Projections, in
Acoustical Holography, P. S. Green ed., vol.
5. , p. 591 (Plenum Press, New York, 1974).
[4] Greenleaf, J. F. and Johnson, S. A.,
Algebraic Reconstruction of Spatial Distri-
butions of Refractive Index and Attenuation
in Tissues from Time-of-Fl ight and Profiles,
in Ultrasonic Tissue Characterisation, M.
Linzer ed.. National Bureau of Standards
Spec. Publ. 453, p. 109 (U.S. Government
Printing Office, Washington, D.C., 1976).
[5] Radon, J., Uber die Bestimmung von Funktionen
durch ihre Integralwerte langs gewisser
Mannigfaltigkeiten. (On the determination
of functions from their integrals along cer-
tain manifolds.) Berichte Sachsische Aka-
demie der Wissenschaften (Leipzig), Mathe-
matische-Physische Klasse 69. 262-277 (1917).
[6] Gordon, R. and Herman, G. T. , Three-dimension-
al reconstruction from projections: a review
of algorithms. International Review of Cytolo-
gy. 38 , 111-151 (1974).
[7] Gordon, R. , A tutorial on ART, IEEE Trans.
Nuc. Sci. . NS-21, 78-93 (1974).
251
CHAPTER 9
SIGNAL PROCESSING AND PATTERN RECOGNITION
253
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed. , National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
A COMPREHENSIVE ULTRASONIC TISSUE ANALYSIS SYSTEM
M. Linzer, S. I. Parks, S. J. Norton,
F. P. Higgins', D. R. Dietz^ and R. W. Shideler
National Measurement Laboratory
National Bureau of Standards
Washington, D.C. 20234
and
T. H. Shawker and J. L. Doppman
Clinical Center
National Institutes of Health
Bethesda, Maryland 20014
A progress report on the development of a comprehensive system for ultrasonic
tissue characterization is presented. Major elements of the program include com-
puterized tomography studies, particularly for breast cancer detection; opto-acoustic
visualization of ultrasonic fields, for testing of new imaging schemes, studies of
propagation through inhomogeneous media, in vitro measurements, and transducer cali-
bration; electronic focusing, especially annular array imaging; sensitivity enhance-
ment, using digital signal averaging and pulse compression techniques; computer and
chirp waveform techniques for compensation of frequency-dependent attenuation; the
SonoChromascope, a digital device for real-time acquisition, processing, and display
of B-scan images; and computer-based image processing.
Key words: Annular array; breast cancer; chirp signals; imaging; opto-acoustic;
pulse compression; sensitivity; signal averaging; signal processing;
tissue characterization; tomography; transducers; ultrasonics.
I. INTRODUCTION
This paper reports on the progress of a Com-
prehensive Ultrasonic Tissue Analysis System
(CUTAS) which is being developed by the Signal
Processing and Imaging Group, Center for Materials
Science, NBS, and clinically-evaluated by the
Department of Diagnostic Radiology, Clinical
Center, NIH. Three principal areas are being
emphasized in this work:
1. Computerized tomography studies, particularly
for breast imaging,
2. Optical visualization of ultrasonic fields.
3. General purpose clinical studies.
II. COMPUTERIZED TOMOGRAPHY STUDIES
Major emphasis to date has been the develop-
ment of backprojection algorithms for reconstruc-
tion of cross-sectional images of reflectivity
using a circular array of transducer elements
enclosing the object. Three basic modes of data
acquisition and image reconstruction were
analyzed (fig. 1): (1) the same element serves
as transmitter and receiver and data is backpro-
jected along circular paths centered at the
'Present address: Western Electric Company,
Princeton, New Jersey
^Present address: McDonnell Douglas Corporation,
St. Louis, Missouri
0 2p/c 0 (P,+p,l/c
Fig. 1. (a) Operation of circular array with
same element serving as transmitter
and receiver. Each point in the
A-scan is the sume of echoes arising
from scatterers along a circular arc
centered at the active element, (b)
Operation of circular array with sepa-
rate transmitter and receiver elements.
Each point in the A-scan is the sum of
echoes arising from points lying along
an elliptical arc whose foci are the
transmitter and receiver.
element; (2) distinct transmitter and receiver
with fixed separation and backprojection along
elliptical paths with the elements at the foci;
and (3) distinct transmitter and receiver with
varying separations and backprojection along
corresponding elliptical paths. Point spread
functions (PSF's) were evaluated for narrowband,
wideband, and an analytically-derived optimum
pulse which yields the best sidelobe response
and a mainlobe width equal to one-third of the
255
1.00
0.75
0.50
1 0.25
-0.25
— - — si'Nyx/y , y\/\,^Nv ^
-4
-2 0 2
r Imm)
Fig. 2. Point spread function obtained by back-
projecting an analytically-derived
optimum pulse along circular paths in
image space.
including that of the transducer face. The ultra-
sonic wavefronts at other planes were also
measured and compared to the calculated values.
Both narrowband (gated-cw) and wideband (pulsed)
waveforms were used to excite the transducer.
Computerized holographic reconstructions of
several model targets were also made. An example
of a reconstructed plane in the field of a pulsed
transducer is shown in figure 3. Other planned
NORMALIZED
DISPLACEMEfjT
AMPLITUtSE
96 mm X 6 4 ps
96 mm X 6,4 ps
wavelength corresponding to the cut-off frequency
of the pulse (fig. 2). When backprojection is
performed along elliptical paths, corresponding
to two separate elements, the mainlobe of the PSF
was shown to be broadened by a factor proportional
to the cosine of half the angle subtending the
two elements at the center of the array. This
behavior places a practical restriction of about
45° on the angular separation of the elements.
Computer simulations confirmed the salient
properties predicted by the analytically-derived
PSF's. The characteristics of the PSF's were
calculated as a function of the number of array
elements, the position of the reflecting point
within the object, and the shape of the pulse.
The backprojection analysis has been extended
to three-dimensional reconstructions (using
spherical transducer arrays) and to incorporate
corrections for velocity inhomogeneities in the
medium. Work in progress includes the develop-
ment of a perturbation approach for correcting
velocity images and algorithms for frequency-
dependent time-gain compensation. A prototype
breast scanner is now under construction and will
be used to evaluate these various developments
in ultrasound tomography.
III. OPTICAL VISUALIZATION OF ULTRASONIC FIELDS
A major laboratory facility for reconstruction
and optical visualization of ultrasonic fields
has been developed. The system is based on the
use of a Michel son interferometric system which
is capable of measuring the displacement of a
thin metallized 150 mm diameter pellicle immersed
in water. Major advances of the NBS development
over previous designs include measurement of both
the amplitude and phase of the displacement, and
absolute calibration of the displacement to
approximately one percent accuracy.
This capability of measuring the complex ultra-
sound field over a large aperture, coupled with
computer acquisition and analysis of the experi-
mental data, has provided us with a powerful
research tool for visualization and reconstruction
of ultrasonic wavefronts. In our initial work,
the ultrasonic field of a transducer was measured
at the plane of the pellicle placed near the
transducer face. The exact (plane wave) solution
was used to reconstruct the field at other planes.
Fig. 3. (a) Computer reconstruction of a propagat
ing wideband pulse at 100 mm (far-field)
from transducer face. Reconstruction was
based on measurement of the time-depend-
ence of the amplitude and phase of the
pulse in a near-field plane, (b) Actual
measurement of pulse at 100 mm plane.
uses of the system include studies of new ultra-
sonic imaging schemes, measurements of ultrasonic
parameters of tissue in vitro, and examination of
ultrasonic propagation through inhomogeneous
media.
IV. GENERAL-PURPOSE CLINICAL STUDIES
A general -purpose clinical scanner is now under
active development (fig. 4). Major components
of the system include annular array focusing for
improving resolution; sensitivity enhancement
techniques using A- and B-scan averaging, chirp-
radar approaches and focusing; first-order
corrections for frequency-dependent attenuation
in the medium; and improved signal processing
approaches combining the real-time acquisition,
processing, and display capability of the Sono-
Chromascope with computer post-processing. To
date, our approach has been to evaluate each
component separately as it is developed. Even-
tually, we expect to integrate these various
elements into a unified facility for clinical
ultrasound examinations.
A. Annular Array
1. Expanding-Aperture Annular Array
A dynamically-focused annular array system for
contact B-scanning has been developed (fig. 5).
The design is based on a constant F-number
approach, whereby, at short focal lengths, the
aperture is expanded in proportion to the focal
length. This approach allows the use of wider
array elements, thus increasing the sensitivity
of the system and reducing radial -mode coupling.
Other major advantages include a substantial
reduction in the delays and refocusing rates
required for the lens synthesis with a correspon-
ding reduction in the electronic complexity of
the system. The initial design employs an array.
256
ULTBAFAST
SIGNAL
ENHANCER
A SCAN DIGITIZATION
AND AVERAGING
ANNULAR
ARRAY
SCANNER
RFSUBSYSTEM
PREPROCESSING
SONOCHROMASCOPE
MULTIPARAMETER
ACQUISITION PROCESSING
AND DISPLAY
NONLINEAR
STUDIES
VIDEOTAPE
A SCAN
RECORDER
Fig. 4. Block diagram of general purpose clinical ultrasound system.
FREQUENCY 25MHi
DIAMETER 3 9 CM
MUMBER OF RINGS 12
Fig. 5. Operation of the expanding-aperture
array with dynamic focusing on receive.
operating at 2.25 MHz, with four annul i active
at the near focal length of 1.5 cm. As the focal
length increases, the array expands to a maximum
of twelve rings, with 4.0 cm outer diameter, for
focal lengths greater than 12 cm. A single,
tapped delay line with 1 \is total duration pro-
vides the time delays for focusing on receive.
A continuously-variable point or line focus is
provided on transmit. Experimental measurements
vides the time delays for focusing on receive.
A continuously-variable point or line focus is
provided on transmit. Experimental measurements
of the focusing properties of the system have been
made. Some of these results are shown in figures
6 and 7. Clinical evaluation of the annular array
scanner is now underway.
Fig. 6. B-scans of AIUM test object with a single
fixed-focus transducer (a) and with the
dynamically-focused annular array (b).
>
1 1 1 —
' A '
-r I 1 '
Uni
:rary
(Arbil
/ ^■'^ \
j mm \
IGNAL
.1 1 1 1 1 1 1
-8 -6 -4
-2 0 2
4 6 8
OFF-AXIS DISTANCE IN FOCAL PLANE (mm)
Fig. 7. Measured response of the annular array
focused at 10 cm on transmit and receive.
2. Wideband Annular Array Response
Theoretical studies of the effect of bandwidth
on the focal plane response of a circular lens
and annular array were carried out. Particular
emphasis was placed on lens systems operating at
approximately 50 percent bandwidth, typical of
those used in ultrasound imaging. An analytical
model of the focal plane response of both the
circular lens and annulus, driven by an impulse,
was developed. The wideband response was then
calculated by convolving the impulse response
with the driving function. For a circular lens,
the beam width in the focal plane, as well as the
position and height of the sidelobes, was analyzed
as a function of bandwidth and aperture weighting.
The wide bandwidth model of an annulus was used to
calculate the response of an annular array. A
detailed comparison was made of this model with
the experimentally-measured response of an array,
operating at 2.25 MHz with 40 percent bandwidth
(fig. 8).
B. Ultrafast Signal Averaging and Pulse
Compression Techniques
A signal averager and pulse compression system
has been developed for sensitivity enhancement
in ultrasonic diagnosis. Potential applications
include the use of transducers which are ineffi-
257
-8 -6 -4 -2 0 2 4 6 8
OFF-AXIS DISTANCE p (mm)
Fig. 9. Ultrasonic reflection from posterior
cortex of tibia; (a) no averaging;
(b) after averaging for scans.
Fig. 8. Field pattern at 10 cm generated by a
0.4 diameter piston transmitter and a
thin, 1.4 cm diameter, annulus receiver.
The solid line is the calculated
response and the dashed line is the
measured response.
cient but otherwise have very desirable features
(e.g. , point and line sources, polymer trans-
ducers, CdS phase-insensitive transducers);
higher frequency operation for improved pattern
recognition and increased resolution; detection
of small reflections; examination of obese
patients; and penetration through skull and bone.
These systems also make it possible to reduce
peak power while keeping average power, and hence
sensitivity, constant.
1 . Signal Averager
The signal averager is capable of real-time
(unbuffered) averaging at 50 MHz rates. To our
knowledge, this device is the fastest digital
averaging device in existence. Major features
include 4K 24-bit words, 12.5 kHz maximum repeti-
tion rate, computer interface, 6-digit cursor
readout of signal amplitude, region of interest
expansion (up to a factor of 16), 3-digit setta-
bility of sample rate, internal/external trigger,
internal delay, segmented memory capability (full,
halves, quadrants, octants), plug-in ADC's (4 bit,
50 MHz; 8 bit, 20 MHz), ADC resolution enhancement
(via_ ordered dither), display normalization and
semi-real time display at high frequencies. The
sensitivity-enhancement capability of the averager
was demonstrated on a PZT line source and on a
number of highly-attenuating biomaterial s, includ-
ing human tibia (fig. 9), and a fiber composite
(fig. 10) used in synthetic implants.
2. Pulse Compression
The pulse compression circuit incorporates a
surface acoustic wave (SAW) "chirp" filter. Pulse
compression ratios of 30:1 and 8:1 have been
obtained in the case of 8 MHz and 3 MHz filter
bandwidths, respectively. An example of both an
expanded and compressed (8:1 compression ratio)
echo from a human heart in vivo is shown in
figure 11. The chirp system was also used to
compensate for frequency-dependent attenuation in
the medium by modulating the amplitude of the
(a)
(b)
Fig. 10. Backwall reflection from 5 cm thick
fiber composite: (a) no averaging;
(b) after averaging for 2^^ scans.
Fig. n. Preamplified echo from a human heart in
vivo. The initial transient is the
clipped transmission pulse; the remaining
echoes are from tissue: (a) expanded
pulse; (b) pulse after 8:1 compression.
expanded stimulus pulse as a function of time.
Since the frequency of this pulse was also time-
dependent, the strongly-attenuated frequencies
were enhanced. This approach has important impli-
cations for improving range resolution and for
tissue characterization and is now under detailed
investigation.
258
C. SonoChromascope
The SonoChromascope (fig. 12) is a state-of-
the-art device for the digital acquisition,
processing, recording, and display of ultrasonic
B-scaii images.
Fig. 12. Clinical ultrasound examination using
the SonoChromascope.
imaging, including electronic focusing and com-
puterized tomography; sensitivity enhancement;
measurement of tissue parameters in vitro;
studies of ultrasound propagation through inhomog-
eneous media; computer and chirp techniques for
compensation of frequency-dependent attenuation;
real-time digital processing and display tech-
niques; and computer-based image processing.
Several of these developments have already been
completed, and, v/here appropriate, are undergoing
clinical testing. Complete descriptions of this
work will be published shortly.
Acquisition algorithms include a choice of
log or linear detection; recording the minimum
and/or maximum echo for each pass through a pixel
(thus, for example, providing a measure of
scattering anisotropy); unconditionally writing
the last value; and summing with normalization
(thereby improving sensitivity). A combination
of up to four different algorithms may be applied
simultaneously to produce different spatially-
congruent images during the same B-scan.
After acquisition, the image or images are
displayed on two color-TV monitors. Images in
complementary colors may be overlaid to permit
visual discrimination of subtle differences. A
variety of thresholding and display modes, aug-
mented by a lightpen, permit semiquantitative
measurements of ultrasound echo intensity. To
use the lightpen, the operator "paints in" the
area of interest. As he does so, the average
image intensity (or intensities) and the area
painted in are displayed on digital readouts.
If more sophisticated processing or storage
is required, the data may be transferred to a
minicomputer. Processed or stored images may be
returned to the SonoChromascope for display.
Spatial and amplitude resolution depend upon
the acquisition mode. For example, in the "un-
conditional write" mode, the word size is 8 bits
and the picture contains 480 x 480 pixels; in
the sum mode, the word size is 14 bits and there
are 480 x 240 pixels.
The SonoChromascope is presently interfaced
to a commercial B-scan system, from which it
receives the rf^or- log -detected A-scan signal
and appropriate information about transducer
position. It is now undergoing clinical evalua-
tion.
V. SUMMA.RY
A major effort is now underway in developing
a comprehensive system for ultrasonic tissue
characterization. The program encompasses
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed.. National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
THEORETICAL ANALYSIS OF INSTANTANEOUS POWER SPECTRA
AS APPLIED TO SPECTRA-COLOR ULTRASONOGRAPHY
W. D. Jennings, E. Holasek, and E. W. Purnell
Department of Surgery, Division of Ophthalmology
School of Medicine
Case Western Reserve University
Cleveland, Ohio 44106, U.S.A.
Spectra-color ultrasonography (SCU), a technique for two-dimensional (B-mode) dis-
play of ultrasonic spectral data, has been analyzed theoretically. The analysis in-
cludes the use of an instantaneous power spectrum calculated from broadband gated echo
spectra produced by an analog spectrum analyzer. The results of the analysis indicate
that additional signal processing factors must be added to the SCU system as it was
originally designed. With the inclusion of these modifications, the SCU scan rep-
resents a true low resolution instantaneous spectral analysis of ultrasonic echo wave-
forms. An experiment was performed to test the theoretical equations we have develop-
ed relating SCU to an instantaneous spectral analysis. The comparison of the SCU
signals and the computed SCU equivalent based on an instantaneous power spectrum is
presented.
Key words: Instantaneous power spectra: Color-coded B-scan; spectra-color ultra-
sonography (SCU); spectrum analysis; ultrasonic spectroscopy.
1. Introduction
Several years ago we developed spectra-color
ultrasonography (SCU) which was originally de-
signed as an approximate spectroscopic technique
to display two-dimensional (B-mode) spectroscopic
information from a sector scanned tissue segment
[l,2]i. Originally, the SCU image was formed by
first passing a broadband B-scan ultrasonic echo
train through a voltage tuned filter set at three
different center frequencies (bandpass positions).
Each of the three resultant filtered B-scans was
then assigned a color (red, green or blue). The
three colored B-scans were then superimposed to
form a single frequency-dependent color coded
B-scan display. The final colored B-mode image
displayed frequency-dependent properties of the
tissue being scanned, though the precise mean-
ing of the display was not apparent.
The subject of this presentation is an analysis
of the SCU process in terms of an instantaneous
power spectrum. An instantaneous power spectrum
is a time dependent spectral decomposition of a
signal. The conventional Fourier Power Spectrum
of a signal is defined by an integral of the
signal over all time, and thus, is independent of
time. An instantaneous power spectrum, however,
is a function of both time and frequency. It is
the dual functionality of instantaneous power
spectra that makes them applicable to signals that
^Figures in brackets indicate literature
references at the end of this paper.
originate from spatially distributed sources with
varying spectral content, such as ultrasonic echo
waveforms.
Several definitions of an instantaneous power
spectrum have been devised. We began our work on
instantaneous power spectra by attempting to adapt
a particular instantaneous power spectrum, the one
defined by Chester Page [3], to spectra-color
ultrasonography. During our investigations, ex-
perimental evidence led us to the conclusion that
for spectra-color ultrasonography to represent a
true instantaneous power spectrum, it is appro-
priate to use Morris Levin's modification of
Page's definition of an instantaneous power spec-
trum, rather than Page's original definition.
This paper is an account of those investigations.
2. Theory
Essential to the analysis of spectra-color
ultrasonography, is the concept of an instantaneous
power spectrum. We began with the approach of
Chester Page [3] which we will briefly review here.
For any time domain signal g(t)2, the running
Fourier transform of g(t) is defined as
t
G^(co) = f g(x)e-J'^^dx . (1)
2ln our case, g(t) is the receiving transducer
voltage corresponding to the instantaneous
acoustic pressure on the transducer face.
261
That is, G^((;)) is the Fourier transform of the
semi-infimte signal g(x) taken from x = -«> to
X = t. Note here that x is a dummy variable, al-
lowing us to use the time variable t as the trail-
ing edge of the gate on g(t). The instantaneous
power spectrum of g(t) can then be defined as
p(t,co) =
3t
|G^(a))
Page also shows that p(t,to) can be represented by
p(t,co) = 2g(t)Re
Jwt
xle-J-^^dx
(3)
By combining the exponential terms, and interchang-
ing the order of integration we get
J.(t) = 2g(t)Re
(2) or
Ji(t) = 2g(t)Re
J j C.(a))g(x)eJ'^'^"'')da)dx
/ g(x)/ C.(a,)ej'^(^-^)da)dx
(9)
(10)
By use of eq. (4), eq. (10) becomes
Using eq. (3) as a starting point, we have ana-
lyzed SCU as follows: In the original SCU implemen-
tation, the signal g(t) was passed through a voltage
tuned filter set at three different bandpass posi-
tions. In the frequency domain, the i**^ filter can
be described by some function C-j(u). In general,
Ci(u)) is a complex function, including phase delays
introduced by the filter as a function of frequency.
The time domain representation of the filter will be
designated by c-j(t), which is also complex in general
The function c-j(t), is the inverse Fourier transform
of C-j(u), and is the impulse response of the filter
J.(t) = 4TTg(t)Re
J g(x)c.(t-
x)dx
(11)
For a filter that
(ci(t) = 0 for t
is causal in the time domain
: 0) , eq. (11) becomes
Ji(t)
4TTg(t)Re
j g(x)c.(t-x)dx
(12)
:.(t) = J-/c.(.)eJ
jwt
dto
(4)
Thus, the filtered signal can be described in the
time domain by
S.(t) = Re[c.(t)*g(t)j
(5)
What we would like to display is the band-limited
instantaneous power spectrum of the signal. That
is, the desired display is
Ji(t) = j p(t,w)C.(aj)dco
(6)
Ji(t) is the instantaneous power (as defined by
Page) of the signal g(t) in the frequency band de-
scribed by C-i(a)). Substituting eq. (3) into eq.
(6) and rearranging we get
J.(t) = 2g(t) j C.(w)Re e^^^^y g(x)e"J^^dx
dw . (7)
If we assume at this point that the filter function
Ci(u)) is a real valued function, rather than com-
plex, we can write
J.(t) = 2g(t)Re
or
J.(t) = 2g(t)Re
/
C.(a3)e
jut
/g(^
x)e '^'^^dxdw
(8)
j j C.(a))eJ'^S(x)e"^'''dxdw
The integral in eq. (12) is the convolution in-
tegral for g(t) and c-j(t). Using the notation of
eq. (5), eq. (12) becomes
J^(t) = 4^g(t)S.(t)
(13)
Thus, we come to the conclusion that for SCU to
represent a true color coded low resolution in-
stantaneous spectral analysis we must modify the
original system in two ways:
(1) The filters used must have filter functions
that are real in the frequency domain, or be
modified so as to be effectively real.
(2) The filtered waveforms must be multiplied
by the broadband signal before video processing
and display.
3. Experimental Approach
We have tested the theory by comparing J-j(t)
calculated by eq. (6) with J-j(t) defined by eq.
(13), which mathematically represents the modified
SCU system.
Calculation of J-j(t) by eq. (6) requires mea-
surement of the filter functions C-j(a)). We used
three fixed filters with 1.6 MHz bandwidths center-
ed at 7, 9, and 11 MHz. The fixed filters re-
placed the voltage tuned filter used in earlier
implementations of spectra-color ultrasonography.
The spectrum of the ultrasonic echo from a glass
block was used as a reference spectrum, and the
spectrum of the signal passed through each filter
was recorded. The filtered spectra and reference
are shown in figure 1. The spectra are plotted
from 5 to 15 MHz. A 5 MHz crystal controlled
oscillator was used to provide a frequency marker
for accurate spectral measurements. All spectra
were recorded by our automated data acquisition
262
wideband
9 10 11 12 13
Frequency (MHz)
14
Fig.
1. Wideband reference spectrum and spectra of
the low, middle and hiqh bandpass filtered
reference signals.
system [4] controlled by a programmable calculator.
A natural sponge was used as the target mate-
rial because of its stability and similarity to
human tissue in acoustical properties. The in-
stantaneous power spectrum, p(t,a)) required by
eq. (6), was calculated by eq. (2). To produce
the semi-infinite signal gt(x) we constructed a
programmable gating system. Using the crystal
controlled 5 MHz source as a clock, the gate could
be automatically varied in length in 200 nano-
second increments. Power spectra |G^((o)|2 of the
signal gated at increasing lengths were recorded.
The value of (3/3t) |Gt(uj)|2 was calculated by sub-
tracting spectra of gated signals of increasing
length to approximate the differential. Ji(t)
was then calculated as the summation of
Ci(io) • (8/9^.) |Gt(a)) |2 over all frequencies to ap-
proximate the integral in eq. (6).
To determine Ji(t) by eq. (13), the filtered
signal, Si(t), was multiplied by the broadband
signal, g(t), using a double balanced modulator.
A broadband pulse amplifier was used to bring the
level of g(t) into the operating range (approxi-
mately one volt) of the modulator. The modulator
provided bipolar multiplication of the two signals,
as shown in figure 2, The multiplied signal for
each of the three bandpass filters is shown in
figure 3.
For each filter, an appropriate precision de-
lay line was added to the broadband signal before
multiplication in the modulator. The delay lines
Fig. 2.
2c-
Analog signal multiplication for the
middle frequency band, a) multiplied
signal, J(t); b) wideband signal, g(t):
c) filtered waveform, Si(t).
3a
3b
u
i
3c
Fig. 3. Analog multiplied signals, J^(t) fot the
three frequency bands, a) high frequency
band; b) middle frequency band; c) low
frequency band.
were necessary to compensate for the group delay
through the filters. Group delays were approxi-
mately 1 microsecond, but were different for each
filter. Thus, the requirement that the filter
functions be real was approximately satisfied.
4. Results and Discussion
Figures 4a, 4b, and 4c, show J-i(t) calculated by
eq. (6), and J-j(t) determined by the signal multi-
plication described in eq. (13) for the three
filters. The time resolution of the calculated
signals is one point every 200 nanoseconds, while
the analog multiplied J^{t) is a continuous func-
tion of time. The calculated Ji(t) represents the
average of J-j(t) over the 200 nanosecond interval.
Noise in the system was reduced by averaging in
the case of the calculated Ji(t). The plots shown
in figure 4, represent an average of 40 scans
through the 5 microsecond section of sponge. Fig-
ure 5 shows the individual scans before averaging,
and figure 6 shows plots of the averaged signals.
The noise becomes progressively higher with deeper
signal penetration, as shown in figure 5, This
effect is due to the internal noise of the spectrum
analyzer. The signal to noise ratio of the spec-
trum analyzer was constant for all signal levels,
since the spectrum analyzer was used in the loga-
rithmic mode. Thus, for higher signal levels
263
4c
[l
1
1
0 1 2 3 4 5
Microseconds
Fig. 4. Comparison of J-j(t) calculated by eq. (6)
(upper trace) with the multiplied J-j(t) de-
fined by eq. (13) (lower trace) for the
three filters, a) high frequency signals;
b) middle frequency signals; c) low fre-
quency signals.
(longer gates) the absolute noise level increased.
This resulted in a decreased signal to noise ratio
for the differential power spectrum of higher level
signals.
It is interesting to note that the time resolu-
tion of the multiplied signal J-j(t) is better than
the resolution of the broadband signal g(t) and
significantly better than the resolution of the
filtered signals, S-j(t) used in the original SCU
implementation (see fig. 2). This is interpreted
as a result of having increased the information
bandwidth of the broadband signal g(t) by the band-
width of the filtered waveform Si(t) in the multi-
plied signal Ji(t).
We can also observe that agreement between the
calculated Ji(t) and the analog multiplied Ji(t)
progressively improves as we go from the high
frequency band to the low frequency band. This
trend is probably due to the fact that the assump-
tion of a real filter function is better justified
for the low frequency filter than for the high
frequency filter. The time delay error which was
approximately constant for the three filters, is
most disruptive to the high bandpass filter, be-
cause of the shorter repeat period of the filtered
si gnal .
The results of this experiment show only fair
agreement between the band limited instantaneous
power calculated by eq. (6), and the band limited
instantaneous power produced according to eq. (13).
This result demands that we reexamine the assump-
tions made in the theoretical analysis. The first
assumption of a real valued filter function is
reasonably valid because of the broadband delay
line compensation for the group delay through the
filters. The appearance of predominately positive
signals in the multiplied waveforms indicates that
the filtered and broadband waveforms are largely
in phase, though the filters have some dispersive
effects. The second assumption was that the filter
is causal in the time domain. This assumption also
appears valid for passive filters. However, close
examination reveals that the two assumptions above
are mutually exclusive. That is, if a filter func-
tion is real valued in the frequency domain, it
must be noncausal in the time domain. This is true
2 3
Microseconds
2 3
Microseconds
Fig. 5. Individual calculator plots of J-j(t) for
40 scans through the sponge, a) high fre-
quency band; b) middle frequency band;
c) low frequency band.
Fig. 6. Calculator plot of the average J^-(t) for
the 40 scans in figure 5. a) high fre-
quency band; b) middle frequency band;
c) low frequency band.
264
because a real valued function, when transformed by
eq. (4), must have a real part (the integral of the
cosine terms) that is an even function of time. An
even function of time must be noncausal. Hence,
the transition from eq. (12) to eq. (13) is not
val id.
5. Time Reversal Symmetry
From the above discussion, it appears impossible
to display Page's instantaneous power spectrum by
simple signal filtering and broadband multiplica-
tion. The restriction to non-causal filters re-
quired by eq. (8) introduces a contribution to the
instantaneous power from the signal at future
times. The time reversal symmetry of the filter
function in the time domain means that eq. (13)
corresponds to an average of Page's instantaneous
power and a similar instantaneous power with the
time scale reversed (i.e., a signal taken from
to time t). An instantaneous power spectrum of
this type has been introduced by Morris Levin [5].
Levin defines an instantaneous power spectrum
analogous to Page's except with the added proper-
ty of time reversal symmetry. According to Levin,
the symmetric iriStantaneous power spectrum is
given by
where
1
2 3t
2^(03)12 - |G^(a))
G^(a))
x)e-J"^dx
(14)
(15)
Ji(t) = f g(t)Re
C^. (a))du
(20)
Again, assuming C^{u) is real valued rather than
complex, eq. (20) becomes
J.(t) = g(t)Re
y G(,o)C.(aOej'"^da)
(21)
The integral in eq. (21) is the inverse Fourier
transform of the product G(o))Ci (u) , and hence be-
comes the convolution of the two corresponding
time domain signals
J.(t) = 2^g(t)Re[g(t)*c.(t)j
or
J.(t) = 2ug(t)S.(t)
(22)
(23)
where S^-(t) is defined by eq. (5). No assumption
of causality of c-j(t) is required in the derivation
of eq. (23) because of the time reversal symmetry
of Levin's instantaneous power spectrum.
Thus the two modifications of the SCU system
stated above (broadband delay and signal multipli-
cation) produce a band limited instantaneous power
corresponding to Levin's instantaneous power spec-
trum, rather than Page's spectrum as initially
intended.
6. Conclusions
and
g:J(co) = / g(x)e-j"^dx . (16)
t
Levin also shows that p^{t,ui) is given by
Ps(t,a>) = g(t)Re [6(^)6^"^*] (17)
where
00
G(w) = f g(t)e"J'"tdt . (18)
G(u) is simply the frequency spectrum of the com-
plete waveform g(t). By analogy to eq. (6) above,
we redefine the band limited instantaneous power
Ji(t) (for Levin's spectrum) to be
The analysis we have presented indicates that
spectra-color ultrasonography should be modified
in two ways. First, the filters used must have
real filter functions in the frequency domain, in-
cluding compensation for the time delays through
the filter. Second, the appropriate signals to
color code and display are the products of the
filtered signals with the broadband echo waveform,
rather than the simple filtered waveforms.
With these two modifications in effect, SCU rep-
resents a true low resolution instantaneous spectral
analysis of ultrasonic echoes corresponding to the
instantaneous power spectrum of Morris Levin. In
addition, the time resolution of the new system
exceeds that of the broadband signal.
Acknowledgments
This work was supported in part by National In-
stitute of Health Grant #EY 00224-15 and The Ohio
Lions Research Foundation.
References
J^it) = / P5(t,a))C.(a))dw (19)
where C^(u)) is defined as before. Substituting
eq. (17) into eq. (19) gives
[1] Holasek, E., Gans, L. A., Purnell, E. W., and
Sokollu, A., A method for spectra-color B-scan
ultrasonography, J. Clinical Ultrasound, 3^,
175-178 (1975).
[2] Purnell, E. W., Sokollu, A., Holasek, E., and
Cappaert, W. E., Clinical spectra-color ultra-
sonography, J. Clinical Ultrasound, 3^, 187-189
(1975).
265
[3] Page, C. H. , Instantaneous power spectra, J_.
Applied Physics, 23, 103-106 (1952).
[4] Holasek, E., Jennings, W. D., Sokollu, A., and
Purnell , E. W., Recognition of Tissue Patterns
by Ultrasonic Spectroscopy, Ultrasonic Sym-
posium Proceedings, 'IEEE Cat. #73 CH 0807-8SU,
73-76 (1973).
[51 Levin, M. J., Instantaneous Spectra and Ambi-
guity Functions, IEEE Transactions Information
Theory, Vol. IT-10, 95-97, (1964).
266
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed., National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C. , 1979).
IDENTIFICATION OF TISSUE PARAMETERS BY DIGITAL PROCESSING OF
REAL-TIME ULTRASONIC CLINICAL CARDIAC DATA
L. Joynt, D. Boyle, H. Rakowski, R. Popp, and W. Beaver
Center for Integrated Electronics in Medicine
and Division of Cardiology
Stanford University
Stanford, California 94305, U.S.A.
A study to assess the feasibility of obtaining diagnostical ly useful tissue charac-
terization information by digitally processing clinical cardiac data is described.
Normal subjects, myocardial infarction, IHSS, and amyloid patients were studied. The
data acquisition system used to record data from a real-time scanner is also described.
Significant changes in the RF signals and frequency spectra as the heart moves were
noted over very short time intervals, indicating the need for a dynamic tissue charac-
terization measure. Wide variation in the spectral characteristics of the signals from
the normal population were found. Behavior of the spectra for the MI patient data were
noted which differentiated them from the other subjects.
Key words: CI inical cardiac data; Fast Fourier Transform; digital processing; frequency
spectra; in vivo; microprocessor-controlled data acquisition; myocardial in-
farction; real-time; tissue characterization; ultrasound diagnosis.
1. Introduction
The possibility of obtaining diagnostic informa-
tion by processing clinical ultrasound data from
cardiac patients is explored in this paper. Lele
et al. [l]i and Yuhas et al. [2] have differentiat-
ed between normal and ischemic or infarcted heart
muscle in vitro on the basis of measurements of
acoustic attenuation. Extension of these methods
to clinical ultrasound examinations is complicated
by factors including cardiac motion, intervening
tissue, and the angle of incidence of the ultra-
sound beam to the interrogated tissue [3]. The
most direct approach to exploring the effects of
these difficulties in the clinical environment is
to interface to an instrument in clinical use.
Coupling a real-time scanner to a high-speed data
acquisition system makes it possible to obtain in-
formation on the dynamic state of the heart. A
preliminary study on a limited number of patients
and normal subjects was done in order to assess the
feasibility of deriving diagnostically useful in-
formation about the type and state of cardiac tis-
sue by digital processing of clinically acquired
data.
2. Data Acquisition
A. The Data Acquisition System
The design of the data collection system for
this project is based on a number of considerations.
^Figures in brackets indicate literature
references at the end of this paper.
In order to eliminate possible information loss due
to the detection process, the rf signal is digitized
for off-line computer processing. Experience with
an earlier version of the data system had shown that
large changes occur in signals recorded at 100 ms
intervals. Accordingly, the current version was
designed to record as continuously as feasible so
that these changes could be observed in more detail.
Thus, large quantities of data must be recorded at
a very high rate. A compromise solution, trading
off the speed of semiconductor memory for the low
cost of tape memory by using both high and low
speed semiconductor memory to buffer the input to a
magnetic tape drive, is used in the present system.
A Biomation 8100 Transient Recorder is used as the
A/D converter and high speed buffer. The data is
then transferred to a large slow speed buffer, con-
sisting of 32 kilobytes of microcomputer memory,
which stores data from a number of scan lines for
subsequent recording on a magnetic tape drive having
a transfer rate of 30 kilobytes/second. This device
provides mass storage and convenient input to a com-
puter for later processing.
The data acquisition performance of the system
is determined by the limitations of the digital hard-
ware and by the characteristics of the scanner.
When the Varian real-time sector scanner is used as
the signal source, data may be acquired from any one
cursor-selected scan line every 600 microseconds or
from the sequence of scan lines comprising a frame
every 30 milliseconds. The rf signal is digitized
at 20 MHz so that about 100 microseconds or 7 cm of
data can be stored in the transient recorder's 2048
word memory. Only those samples actually correspond-
ing to the area of interest are transferred to the
microcomputer memory buffer. The total time to re-
267
cord data from one scan line is under 2 milliseconds.
Thus, a substantial portion of a frame can be digi-
tized at real-time rates. The system also includes
the option to record a sequence of lines at 30 ms
intervals in order to cover a complete cardiac cycle
in one buffer. When the data buffer is filled the
recording process is interrupted for a few seconds
while the data is transferred to magnetic tape.
Data acquisition is triggered at a pre-selected
point of the cardiac cycle. A simultaneous Polaroid
picture can be obtained. If more than one buffer is
required, subsequent buffers also begin at the select-
ed point of the cardiac cycle.
CRT
TERMINAL
ULTRASONIC
SCANNER
MICRO-COMPUTER
TRANSFER AT 30 KB/s
INTERFACE
AND
DMA CONTROLLER
9-TRACK MAGNETIC
TAPE
TRANSFER AT 2 MB/s
SAMPLED AT 20 MB/s
BIOMATION 8100
TRANSIENT RECORDER
MAGUS SYSTEM
- CROSS-ASSEMBLER
- SIGNAL PROCESSING
- GRAPHIC OUTPUT
Fig. 1. Block diagram of the data acquisition
system. The RF signal is sampled at 20
MHz, filling up the transient recorder
buffer at 20 Megabytes/second, trans-
ferred to microcomputer memory at 2
Megabytes/second, and transferred to
magnetic tape at 30 kilobytes/second.
The data system has been designed to be easy to
use in a clinical environment. The operator can
interact with the microcomputer software system to
specify the data-taking procedure. Information
about the portion of the frame to be digitized is
stored in a table and written on the tape along with
the data for use in processing. Comments can be ad-
ded from the operator's CRT terminal to record ad-
ditional information about the tissue under study.
B. Data Acquisition Procedure
A variety of patients with different myocardial
pathologies were chosen to exhibit various degrees
of focal or diffuse replacement of normal myocardium.
Included were patients with transmural myocardial
infarction (2), idiopathic hypertrophic subaortic
stenosis (IHSS) (3), and cardiac amyloidosis (1).
Three normal subjects were studied as controls.
Consideration of changes in the echoes due to
motion and muscle contraction during the cardiac
cycle is particularly important in a clinical study
of cardiac patients. In order to study these changes
single scan lines through the septum and posterior
wall of the left ventricle were recorded at 2 ms and
30 ms intervals. Data-taking was synchronized with
the patient's electrocardiogram to commence at end
systole and end diastole. In the patients with
myocardial infarction, the area of infarction was
identified as an akinetic left ventricular segment
corresponding to the region denoted by Q waves on
the electrocardiogram. Data was collected from this
segment and compared to normally contracting seg-
ments in the same patient. In the patients with
IHSS, the region of asymmetric septal hypertrophy
could easily be identified and often has an abnormal
ground-glass appearance when viewed in the real-time
image [4]. Again, comparison was made with adjacent
areas of normal myocardial appearance and thickness
and with the posterior left ventricular wall. In
the patient with diffuse myocardial involvement with
cardiac amyloidosis, representative areas of inter-
ventricular septum and posterior left ventricular
free wall were studied.
3. Data Analysis
The recorded ultrasonic data was processed on an
HP21MX minicomputer. Fourier transforms of each re-
corded data trace were computed using the Fast
Fourier Transform (FFT) algorithm. By taking the
Fourier transform of a single echo, one obtains in-
formation about the overlying tissue through which
the acoustic pulse has passed, while the Fourier
transform over an interval containing several echoes
gives information about the spatial structure of the
tissue through interference effects. Periodicities
of the spectral peaks due to constructive inter-
ference between scatterers spaced by a distance d
will be spaced by Af = c/2d, where c is the acoustic
velocity.
The system impulse response of the Varian scanner
is illustrated in figure 2. The limited bandwidth,
from about 1.8 to 2.9 MHz, has implications for
both kinds of tissue information sought. In vitro
studies by other workers [1-8] have shown that dif-
ferences in frequency-dependent attenuation between
normal and ischemic or infarcted heart tissue are
small over this range. Only periodicities cor-
responding to spatial separations larger than .75
mm will appear in this narrow bandwidth. Thus the
available tissue characterization information would
be derived from observing shifts in the overall
shape of the spectrum.
A primary interest was to note the effects of
cardiac motion on the spectra of returned echoes.
We tried to assess changes in the spectra as a
function of time in order to see possible differ-
ences in the spectra corresponding to the anatomic
differences between the normal and diseased heart
muscle. Figures 3 through 7 show examples of the
data processing sequence for the long axis view of
the septum at end diastole for one patient in each
of the groups studied.
4. Results
Substantial changes in the RF signals returned
from the heart and their spectra were observed over >
time intervals short compared to the cardiac cycle.
Data recorded at 2 ms intervals appeared to ade-
quately represent the dynamically varying signal,
whereas data recorded at 30 ms changed abruptly and
apparently randomly between traces. Figures 8 and
9 demonstrate this effect. A dynamic characteriza-
tion of heart tissue is probably indicated since
parameter measurements derived from the frequency
268
02468 10 0 1 2 3 4 5
MICROSECONDS FREQUENCY (MHz)
Fig. 2. a) Impulse response of Varian Real-time Sector Scanner measured by recording a single
echo returned from the front side of a Lucite block in a water tank. This echo was
digitized at 100 MHz. b) 1024 point FFT of impulse response.
MICROSECONDS FREQUENCY (MHz)
Fig. 3. a) Photograph of the scanner image. The rf signal was recorded along the scan line
indicated by the bright cursor line on the image, b) The recorded rf trace is synchronized
to the desired point in ECG and to the picture of the scanner image. The operator chooses
the origin of the time scale. Here, echoes from the septum were recorded, c) 512 point FFT.
The portion of the rf trace used in the FFT is outlined in figure b.
spectrum at one instant of time may not adequately
describe the same tissue 30 milliseconds later.
The first, second, and third central moments
were used to characterize the spectrum of each tis-
sue sample, permitting study of changes in the shape
of the spectrum. Differential frequency-dependent
attenuation, as indicated by a shift in the first
moment, for normal versus damaged myocardium was not
observed. In fact, attenuation/centimeter appeared
to be constant across the frequency range involved.
It is likely that frequency dependent effects are
overshadowed by the effects of imaging a tissue
volume deep within the body, due to overlying tis-
.sue, diffraction spreading of the beam, and orienta-
tion of the tissue to the ultrasound beam. There
was a wide range of normal spectral moments which
overlapped those of the diseased patients. Disap-
pointingly, the IHSS tissue, which presents a dif-
ferent appearance on the scanner image, was not
distinguishable from the normal by the shape of the
Fourier spectra. A comparison of normal and IHSS
tissue spectra is depicted in figure 10. The myo-
cardial infarction (MI) patients were set apart
from the others by much less fluctuation in their
spectral mean and variance curves (fig. 11). This
implies that the tissue is moving or contracting
less than normal tissue which correlates well with
the observed image on the real-time display. There
269
(c)
0 10 20 30 40 0 1 2 3 4 5
MICROSECONDS FREQUENCY (MHz)
Fig. 4. a) Photograph of scanner image, b) Recorded rf signal from selected scan line, c) 512 point FFT
AMYLOID
(b)
(c)
Mil*-
10 20 30
MICROSECONDS
40 0 1 2 3 4
FREQUENCY (MHz)
Fig. 5. a) Photograph of scanner image, b) Recorded rf signal from selected scan line, c) 512 point FFT
(a)
NORMAL A
10 20 30
MICROSECONDS
40 0
12 3 4
FREQUENCY (MHz)
Fig. 6. a) Photograph of scanner image, b) Recorded rf signal from selected scan line, c) 512 point FFT
270
(a1
NORMAL
(b)
0
10
20 30
MICROSECONDS
40
Fig. 7.
2 3
FREQUENCY (MHz)
a) Photograph of sccinner image, b) Recorded rf signal from selected scan line, c) 512 point FFT.
Fig. 8.
10 20 30
MICROSECONDS
0 12 3
FREQUENCY (MHz)
a) Sequence of rf echo traces from normal septum. Signals were recorded
at 2 ms intervals. Bottom trace occurred first in time, b) FFT spectra
of signals in a.
was also a spectral shift of approximately 100 kHz
in the means of the spectra between the long axis
and short axis views for the MI data. This is like-
ly due to a more tangential view of the tissue in
short axis, demonstrating the substantial effect of
orientation to the ultrasound beam.
5. Conclusions
Straightforward application of methods [1,2,8]
which have differentiated between normal and ische-
mic or infarcted myocardium in vitro to in vivo
data derived from current clinical instruments is
not likely to be successful due to the triple
whammy of limited bandwidth, overlying tissue, and
cardiac motion. Spectrum skewing due to broad fre-
quency interference effects and analysis of the time
sequence of spectral changes may show some utility.
The substantial changes in the rf and frequency
spectra over very short time intervals lead to
speculation that they reflect the state of contrac-
tion of cardiac muscle as well as changes in geome-
try. The data acquisition system described in this
paper represents a significant advance in the state
271
2.32 MHz
2.30 MHz
2.25 MHz
Fig. 9. a) Four frequency components of the FFT of normal septum data recorded at 2 ms intervals.
Each component is plotted as a function of time, b) Same four frequency components plotted
as a function of time for data recorded at 30 ms intervals.
2.60-
2.40
2.20
MEAN
VARIANCE
42
MILLISECONDS
Fig. 10. Comparison of mean and variance of spectra of normal and IHSS spetal tissue.
Moments are plotted as a function, 2 ms per point.
of the art of the digital recording of ultrasound
rf in terms of flexibility and ease of use as well
as in the kind of data it makes available.
Acknowledgment
This work was supported by the U.S. Department
of Health, Education, and Welfare under grant GM-
17940.
References
[1] Lele, P. P., Mansfield, A. B., Murphy, A. I.,
Namery, J., and Senapti , N., Tissue Charac-
terization by Ultrasonic Frequency-Dependent
Attenuation and Scattering, in Ultrasonic Tis-
sue Characterization, M. Linzer, ed.. National
Bureau of Standards Spec. Publ . 453, pp. 167-
196 (U.S. Government Printing Office, Washing-
272
LONG AXIS
MYOCARDIAL INFARCTION A
SHORT AXIS
28 42
MILLISECONDS
LONG AXIS
MYOCARDIAL INFARCTION B
SHORT AXIS
Fig. 11. Means of the spectra of data from two infarcted spetums. Data is plotted
at 2 millisecond/point for the long and short axis views for both patients.
ton, D.C., 1976). [6]
Yuhas, D. E., Mimb, J. N., Miller, J. G., Wiess,
A. N. , and Sobel , B. E., Changes in Ultrasonic
Attenuation Indicative of Regional Myocardial [7]
Infarction in Ultrasound in Medicine, D. White,
ed. , Vol. 3 (Plenum Press, New York, 1977).
[8]
Chivers, R. C. and Hill, C. R. , A spectral ap-
proach to ultrasonic scattering from human tis-
sue: methods, objective and backscatteri ng
measurements, Phys. Med. Biol . 20 (5), 799-815
(1975). [9]
Rossen, R. M. , Goodman, D. J., Ingam, R. E.,
and Popp, R. L., Echocardiographi c criteria in
the diagnosis of idiopathic hypertrophic sub-
aortic stenosis, Circulation 50, 747-751 (Oct.
1974). [10]
Cooley, J. W. and Tukey, J. W., An algorithm
for the machine calculation of complex Fourier
series. Math. Computation 19, 297-301 (1965).
Oppenheim, A. V. and Schaefer, R. V., Digital
Signal Processing (Prentice-Hall, Inc., Engle-
wood Cliffs, N. J., 1975).
Crawford, F., Waves (McGraw-Hill, New York,
1968) .
Namery, J. and Lele, P. P., Ultrasonic Detec-
tion of Myocardial Infarction in Dog, in
Proceedings of 1974 Ultrasonic Symposium, p.
491 (IEEE Cat. No. 72 CHO 7088 SU, 1974).
Mohammed, A. and Smith, R. G., Data Windowing
in Spectral Analysis, DREA Report 75/2, De-
fense Research Establishment Atlantic, Dart-
mouth, N. S., Research and Development Branch,
Department of National Defense, Canada.
Fields, S. and Dunn, F., Correlation of echo-
graphic visualizability of tissue with bio-
logical composition and physiological state,
J. Acoust. Soc. Am. 54, 809-811 (1973).
273
i'
f
fl
I
p
il
pi
fi
ii
tt
fl
Ii
ii
SI
II
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed. , National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
DYNAMIC AUTOCORRELATION ANALYSIS OF A-SCANS IN VIVO
J. C. Gore,^ S. Leeman,^ C. Metreweli,^ N. J. Plessner,^
and K. Willsoni
^Department of Medical Physics
2 Department of Diagnostic Radiology
Royal Postgraduate Medical School
Hammersmith Hospital, London, England
A realistic tissue model has been analysed to indicate the information that may be
derived from autocorrelation studies of in vivo A-scan echograms, as well as to show
some of the limitations of such techniques. Although many of these difficulties are
not easily overcome when considering single, or even averaged, realisations of the
autocorrelation function (ACF), an analysis of the time course of the ACF is less
subject to such objections.
Several potentially useful clinical applications of such temporal changes are being
investigated. These include measurements of echoes from within heart muscle through-
out the cardiac cycle as a possible indicator of cardiac disease; similar variations
in echoes from the stomach wall during gastric emptying are demonstrated, and a relation-
ship to contractile state is postulated.
Results are also presented which indicate that the perfusion of tissues with blood
may be assessed by this technique, and, on a different physiological time scale, changes
in the placenta throughout pregnancy have been investigated. The technique is being
extended to the study of the response of malignant tumours to treatment.
Key words: A-scan; correlation analysis; temporal changes.
1. Introduction
Tissue characterisation may be defined as the
attempt to establish what quantitative parameters,
other than range information, may be extracted
from ultrasound probing of human tissue; the
quantities derived should reflect intrinsic tissue
properties and in particular its physiological
state, be useful diagnostically, or for pre- and
post-operative assessment, or for monitoring
response to treatment. The attempt is not neces-
sarily restricted to an ultrasound technique, but
this modality is particularly attractive, not only
for the usual reasons, but also because it pro-
vides the facility for receiving and analysing a
signal from a reasonably localised region of tis-
sue, as well as providing images to enable that
region to be accurately placed. Clearly, only
physical properties of tissue, or their possible
change with time, are amenable to detection--often
micro-structural features are of interest, but
1 other possibilites (e.g. , temperature measurement)
are certainly not excluded.
Isolated echo structure may be probed in tissue
' characterisation analyses [l-3]3, even though the
extracting of purely physical, as opposed to geo-
metrical, data is more complicated than may appear
at first sight [3]. Here we will restrict our-
selves, as have many others, to the scattered low-
level signals from a region of tissue, and perform
what may be termed tissue characterisation by
"echo ensemble" analysis. The extent to which
these two methods are complementary is not known
at present, but it may be noted that both are im-
plicit in qualitative attempts to extract diagnos-
tically useful data from grey-scale images.
2. Some Theoretical Ideas
A conceptual basis for the empirical approach
may be laid from theoretical considerations, just
as the impedance mismatch concept simplifies the
interpretation of much of conventional ultrasound
scanning.
The tissue model adopted here comprises an in-
^Figures in brackets indicate literature
references at the end of this paper.
275
homogeneous medium, with local density and com-
pressibility fluctuations generating the ultra-
sound scattering. In this context, the aim of tis-
sue characterisation by echo ensemble analysis is
to uncover the three-dimensional spatial, and even
temporal, distribution of these fluctuations; while
the hope of the technique is that all, or preferab-
ly some, of the information is sufficient to estab-
lish unambiguously the state of the tissue.
A convenient starting point is the wave equa-
tion for ultrasound propagation in a medium with
stationary inhomogeneities of the type described
above [4].
^^P-E^0=H^T(r)0+V,[p(r)Vp] (1)
where p{r_,t) is the acoustic pressure at location
£ and time t;
v{r) is the density fluctuation,
y = (p - Pq)/p
with po the mean, and p(r) the exact local density;
y(r) is the compressibility fluctuation,
Y = (k - K^)/K^
with Kp the mean, and K{rJ the exact local com-
press ibi 1 ity.
Cq is the mean velocity,
'I - (Vo)-^ •
The right-hand side of eq. (1) is conventional-
ly called the source term, since it is easily de-
monstrated [5] to be a "source" of waves, and it
describes the interaction of the medium with the
ultrasound field. The source term depends on both
the fluctuations in the medium and the local, in-
stantaneous value of the pressure wave, as it must
do if it is to describe a scattering situation.
The tissue model described by eq. (1) may be shown
[5], to contain that of Waag and Lerner [6] as a
special case. Since the scattering is known to be
weak, both y and y may be taken to be small quan-
tities. Any scattering calculation requires the
initial conditions to be clearly specified, and
we have adopted an initial pul se pin i ncident
along the Z-direction, of the form:
p^^(r,t) = a(z - CQt)b(h)exp|i(k^z - co^t)} (2)
where a is the (axial) pulse shape, b is the beam
profile, h is the spatial coordinate transverse to
Z, ko is the carrier wave vector amplitude, and
'"0 ~ Coko- Eq. (2) is a good approximation to
pulses from diagnostic machines over much of the
range of interest.
For weak scattering, and for weakly focused in-
cident fields, the far-field final result [7] shows
that the back-scattered pressure amplitude is
generated by the pulse-smoothed density/compressi-
bility fluctuations in the medium. The power
spectrum, of the backscattered echoes (with bound-
ary effects neglected), is given by:
|p^(.)|....||A(^.ko)p |r(|)p (3)
where A is the Fourier transform of the axial
pulse shape, u'* represents a "Rayleigh" factor,
and the effect of the medium is contained only in
the structure factor.
oo 211
<V> = f f dehb(h)y(h,e,z)
0 0
and similarly for r . 6 is the angle variable in
the cylindrical polar coordinates (h,e,Z) used in
the calculation. The Z-integration may be extend-
ed to infinite limits, since the fluctuations are
presumed to vanish outside the (finite) scattering
volume. The structure factor r is the spatial
frequency spectrum of the beam-profile smoothed
tissue structure, and it is seen that, for the
backscattering case, the density and compressibili-
ty fluctuations contribute equally, but with dif-
ferent phase. At other scattering angles the two
types of fluctuations will contribute with differ-
ently varying magnitudes, and they may be regard-
ed as two different types of scattering element:
the one (compressibility) scattering as an acous-
tic monopole, and the other (density) behaving as
an acoustic dipole. This behaviour may be deduced
from the form of the source term in eq. (1), or
more directly, from straightforward physical argu-
ments [5]. Since the relative importance of the
two types of scattering element varies with direc-
tion, different observers employing a fixed-angle
scattering technique may obtain apparently con-
flicting results from the same tissue. In general,
at least six independent one-dimensional experi-
mental functions would have to be measured in
order to reconstruct both y and u from the data,
and since this would seem to be impracticable, it
is emphasized that, for the characterization of
anisotropic tissues by fewer independent func-
tions, it is essential to specify internal tis-
sue landmarks with respect to which scanning
planes and directions may be fixed. If not, in-
tercomparison of results between different work-
ers, and the repeatability of a single investi-
gator's findings, cannot be established.
Quite apart from difficulties inherent in
trying to quantitate the effect of overlying tis-
sues and interfaces, the above calculation shows
that pulse shape, beam profile, and carrier fre-
quency are important quantities which fundamental-
ly influence the scattered pressure power spec-
trum, and which must be specified or, better, de-
convoluted from the data. Moreover, in regions
where the incident pulse is sharply focused
(|vp^p| large), scattering from the density fluc-
tuations becomes more prominent; this may well
prescribe a suitable technique for mapping the
density and compressibility scattering elements
individually, if so desired. In addition, the
measured backscattering is further modified by
the transfer properties of the receiver [8], but
276
all these "system" artifacts are measurable, and
should be prescribed, if not specifically allowed
for.
Pulse smoothing implies a certain loss of fine
detail: scattering experiments do not indicate
the tissue micro-structure (<<A) as such. It is
quite clear from the above that measured pressure
power spectra from the same tissue may differ with
different transducer apertures and excitations,
but it is also worth emphasizing that, in prin-
ciple, the same backscattered power spectrum may
be seen from di ff erent tissues, with different
density and compressibility fluctuations, provid-
ed that Ty-ry remains the same. This merely under-
lines the fact that, even for isotropic media, a
single backscattering experiment is in general in-
sufficient to characterize tissue unambiguously.
Fortunately, in practice, the situation may be
less complicated as there is some indication that,
for most tissues, only the compressibility (i.e.,
elasticity) fluctuations are of importance for
the production of echoes [9].
pen (SAC Graf Pen), interfaced to a HP 2100A com-
puter via a Tektronix 4010 visual display unit.
With this arrangement effective sampling rates of
up to 15 MHz are achieved, although at present it
is more usual to sample echoes from a region
'V 2 cm in extent at a rate of Ih MHz, consistent
with providing estimates of the ACF at lag inter-
vals corresponding to 100 pm of tissue.
The discrete, normalised ACF, cj, is estimated
from N echo samples. Pi, using the unbiased esti-
mator [10],
r=— Vp-P P -P
J N 1 i-,M i+j i+j,M
where
1 i+M
^•,M = (2M + 1) .^^ Pi
3. Method
As indicated in eq. (3), by measuring the power
spectrum of echoes, the structure factor, r, may
in principle be determined over a limited band of
spatial frequencies to within an accuracy set only
by noise and the uncertainty in knowledge of |A|.
However, we have chosen to explore the use of the
time-domain autocorrelation function (ACF) of
echo amplitudes; as well as being easier to cal-
culate, the ACF contains all the information in-
cluded in the power spectrum, and although the
errors inherent in its estimation differ from
those in power spectrum calculations, it seems
likely that on-line ACF measurement may be at
least as suitable for "in vivo" tissue characteri-
zation in real time as Fourier methods. The de-
sired autocorrelation function is given by
c(t) = / p(t)p(t + T)dt
where p(t) = the backscattered acoustic pressure
amplitude at time t.
From eq. (3), it can be shown that
c(t) - 6(t)*?(t)
where * denotes convolu^tion,_
6(t) = the ACF of <y> - <y>
and is an indicator of tissue structure.
5(t) = the ACF of the second derivative of the
impulse response of the pulse-echo sys-
tem [8].
The convolution with the system function, limits
the information available unless steps are taken to
remove its influence, but such further processing
has not yet been attempted by us.
The equipment used in these studies is a Nuclear
Enterprises NE 4102 Diasonograph , operated with
weakly focused transducers centred on 2.5 MHz, with
total system bandwidth approximately W MHz. Sig-
nals from the A-scan receiver are at present re-
corded photographically from an oscilloscope, and
the echo envelope is then digitised with a sonic
is the running mean of the echo amplitude. The
moving average is calculated over a distance that
is sufficiently large not to influence the fine
structure of the ACF estimate but which filters
spurious low frequency fluctuations and strengthens
stationari ty. Such detrending is performed typical-
ly over intervals of ■v 1 cm.
4. Data Analysis
In general, the entire ACF must be considered
when characterizing a single train of echoes, but
one or more particular features of each ACF may
be extracted in order to summarise quantitatively
any given set of echo samples. For example, the
positions (and to a lesser extent the shapes) of
peaks within the ACF indicate distances over which
there exists structural coherence; prominent and
regular peaks indicate a regular arrangement of
scattering sites. The mean number of crossings,
moments of the function, or curve fitting tech-
niques using small numbers of parameters, may each
be used to describe, concisely, essential details
of an estimated ACF, but the method of feature ex-
traction chosen will depend on the particular ap-
pl ication.
In order to overcome some of the objections
that have been indicated above against the use of
single realisations of echo characteristics in
one dimension, the analysis has been restricted
to situations where the tissue investigated has
been found to be isotropic, or where only the
changes with time of echo characteristics are
deemed to be clinically useful. In this way, by
following the time course of echo features, each
tissue acts as its own internal standard and less
reliance need be placed on the values of individu-
al, single measurements. Time-plots of echo
features such as those described above depict
changes in scattering structure, whilst polar
phase diagrams are useful when the change is
periodic. Cross correlation of the ACFs realized
at different times are calculated to yield infor-
mat'lon about changes, relative to a chosen ref-
erence time, and examples are given below.
In analysing echoes in this fashion, the ACF
features to be emphasized have often to be chosen
arbitrarily, and caution has to be exercised when
attaching significance to any particular one.
277
There are some reliable guidelines that may be
derived bv standard techniques to suggest which
ACF values may be regarded as spurious, but a
final decision as to the significance (or other-
wise) of any derived result can be made only on
the basis of clinical, in vivo investigations.
However, as an interim indicator of our aware-
ness of artifactual correlations, the results
given below are marked with the 95 percent con-
fidence limits (a) for an autocorrelation func-
tion generated with N data points from a random
process [11],
5. Results
There are several clinical problems which may
be explored by using temporal characteristics of
the ACFs of echoes to indicate changes within tis-
sues and three classes may be distinguished.
First, there are cyclic events in which echo
characteristics alter in synchronism with some
physiologically significant, repetitive process,
such as respiration, or with the cardiac cycle,
whose phase is easily monitored on the electro-
cardiograph (ECG). It is, for example, reasonable
to suppose that myocardial activity will be asso-
ciated with a redistribution of scattering centres
in the tissue, since it necessarily involves
changes in both its elastic state and density.
The time course of scattered echoes from within
the left ventricular posterior wall is easily fol-
lowed in the conventional mitral valve echogram
scanning direction. Figure 1 depicts the ACFs
measured from a normal heart in end-systole and
end-diastole, showing typical changes from one
phase to the other. The interpretation of this in
MM
Fig. 1. The ACFs of echoes from the left ventricu-
lar posterior wall of a normal heart, mea-
sured in end-systole (S) and end-diastole
(D).
terms of alterations in muscle fibre dimensions or
elastic state remains to be confirmed, whilst the
degree of change in disease remains to be estab-
lished, but the possibility clearly exists of mea-
suring in vivo characteristics which may be relat-
ed to myocardial contractility.
The flow of blood in the systemic circulation
is also directly coupled with cardiac events.
Following the notion that the perfusion of cer-
tain organs with blood proceeds in synchronism
with the heart's action, the possibility of identi-
fying poorly perfused transplanted kidneys is
under investigation. Figure 2 indicates how
echoes from the anterior region of the upper pole
of a viable kidney, investigated some months after
successful transplantation, change with time. The
obvious changes between phases, and the reversion
in the subsequent cycle, is suggestive of a cyclic,
cardiac-synchronised structural reorganisation
within the kidney. This approach to the quantita-
tion of echo characteristics may also be applica-
ble to other organs, such as the liver and pla-
centa, and possibly to tumours.
Fig. 2. Characterisation of echoes from cortex of
transplanted kidney at different times.
Three autocorrelations were measured, at
(cardiac) end-diastole, end-systole, and
the subsequent end-diastole. Curve A is
the cross-correlation of the first diastol-
ic ACF with itself, curve B is the cross-
correlation of the first diastolic and
systolic ACFs, whilst curve C is the cross-
correlation of the two diastolic functions,
and shows a reversion back to the original
state in synchronism with the heart.
A different cyclic process with less regular
period is the muscular activity associated with
gastric emptying. The contraction of the stomach
wall in normal digestion (which may be monitored
over cycles with periods of approximately 20
seconds with conventional M- and B-scan tech-
niques) [12] produces changes in echo charac-
teristics not dissimilar to those associated with
myocardial contraction, and this lends support to
our interpretation of the origin of heart wall
echo changes. Thus, a similar model, viz. , con-
traction of muscle elements causing changes to
their acoustic scattering properties, may be em-
ployed to interpret the results of both cardiac
and gastric measurements. ACFs of echoes from
the anterior stomach wall are shown in figure 3,
and cross correlations of the ACFs from different
phases in figure 4: the position of the first
principal minimum is plotted as a rotating phasor
in figure 5.
278
Fig. 3. The ACFs of echoes from the anterior
stomach wall, contracted (C) and relaxed
(R).
Fig. 5. Phasor diagram showing locus of the posi-
tion of the first minimum of the function
produced by cross-correlating the ACFs of
echoes from stomach wall measured at dif-
ferent times (as in fig. 5). Total cycle
time 20 seconds. The radius of the
circle corresponds to a correlation lag
of 4.8 mm.
.5-
MM
Fig. 4. Characteristics of echoes from stomach
wall at different times. The ACF at time
0 is cross-correlated with itself (A), with
the ACF at a time 4 seconds later when
the stomach is contracted (B), and with
the ACF at a time 8 seconds after con-
traction, when the muscle has again re-
laxed (C). These events were chosen by
reference to the stomach M-scan. Note
the general shape of the cross-correlation
function does not change but the first
minimum moves with contraction.
A second class of time change being investigat-
ed is the alteration in scattering characteristics
of tissues with age or maturity, and of particular
interest are changes in the placenta throughout
pregnancy. Ageing placentas present different grey
scale B-scan appearances: cross-correlation of ACFs
obtained at different stages of normal pregnancy
are shown in figure 6. This technique may also
reveal differences between normal and insufficient
placentas of the same maturity.
A third class of application is the monitoring
of structural changes in tumours during and after
treatment. Such changes have already been observ-
ed in B-scan images of irradiated malignant tumours
and in vivo ultrasonic tissue characterization
using the techniques outlined above may provide a
clinically useful method of assessing tumour kine-
tics [13].
.5-
1 ' \
\ f\ \
\
i\
\\ .7 \ / ,
V 2/ \. /6
V // \ /
V ' MM
\* / / V R 7
■ 1
1 '/ \
C/
_i — / , \ -usy V'v
\ / 2 \ / 4 •'.^ / \ '
1/ \ 1 ^<k»'' \
Fig. 6. Characteristics of echoes from placenta
at different stages of pregnancy. ACFs
were obtained at 15, 24, 38 and 41 weeks,
and each of these ACFs was cross-correlat-
ed with the 15 week ACF to produce curves
A, B, C and D respectively. There appears
to be little change between 15 and 24
weeks, or between 38 and 41 weeks, where-
as there is a clear change between 24 and
38 weeks.
279
6. Conclusions
The use of single autocorrelation functions to
characterize complex, three-dimensional scattering
media may be severely limited because different
tissues may realise similar results. Many of the
objections raised against the use of one-dimen-
sional techniques, however, are overcome if the
time course of events is followed and only changes
in echo characteristics are regarded as signifi-
cant. Whilst the ACF sunsnarises the information
contained in a single train of echoes, the cross
correlation of sequentially produced ACFs sum-
marises dynamic features of scattering. New diag-
nostic information may possibly be derived from
a study of cyclic changes in gastric and cardiac
muscle, and in other organs blood perfusion may
be assessed. Changes in the nature of tissues
with age or treatment may have important signifi-
cance which, as yet, we are unable to evaluate.
A limited number of patients and volunteers
have so far been investigated with this "dynamic"
technique, and it is impossible to unequivocally
claim success in any particular application; but
there is equally no cause to reject the original
concept that echo ACFs reveal changes in the
physiological state of tissues. The influence of
factors such as tissue movement or anisotropy and
the relative merits of different transducers and
methods of signal processing will have to be care-
fully evaluated, and the optimal choice of echo
ACF features to be extracted remains uncertain.
Despite the problems, there is every reason to
hope that dynamic tissue characterization is an
appropriate path to obtaining information of
genuine clinical utility.
Acknowledgements
We gratefully acknowledge Mr. George Hooker's
help with the analysis of data, and thank Dr.
E. W. Emery for his useful discussions and com^
ments. S. L. is in receipt of a Wellcome Fellow-
ship.
References
[1] Lizzi, F., Katz, L., St. Louis, L., and
Coleman, D. J., Applications of Spectral
Analysis in Medical Ultrasonography,
Ultrasonics, 77-80, March 1976.
[2] Trier, H. G., Decker, D., Reuter, R. ,
Epple, E., Lepper, R. D., and Nagel , M. ,
Frequency-modulated Portions of the Time-
amplitude Ultrasonogram of Models, in
Proceedings of the Second European Congress
on Ultrasonics in Medicine, pp. 121-128
(Excerpta Medica, Amsterdam-Oxford, 1975).
[3] Gore, J. C. and Leeman, S., Echo structure
in medical ultrasonic pulse-echo scanning,
Phys. Med. Biol. 22^ (3), 431-443 (1977).
[4] Morse, P. M. and Ingard, K. N., Theoretical
Acoustics (McGraw-Hill, New York, 1968).
[5] Leeman, S. , Simple Physical Ideas on Ultra-
sound Pulse Scattering from Tissue, R.P.M.S.
Medical Physics Reports, US76/3 (1976).
(Available by request only)
[6] Waag, R. C. and Lerner, R. M., in Ultra-
sonics Symposium Proceedings, I . E . E . E .
Cat. 73, CHO 807-850.
[7] Gore, J. C. and Leeman, S., Ultrasonic back-
scattering from human tissue: A realistic
model, Phys. Med. Biol. 22 (2), 317-326
(1977).
[8] Gore, J. C. and Leeman, S., New Criteria
for the Assessment of the Resolution of
Ultrasonic Scanners, in Ultrasonics in
Medicine, E. Kazner, M. de Vlieger, H. R.
Muller, and V. R. McCready, eds., pp. 197-
203 (Excerpta Medica, Amsterdam, 1975).
[9] Fields, S. and Dunn, F. , J. Acoust. Soc.
Am. 54, 809-813 (1973).
[10] Jenkins, G. M. and Watts, D. G., Spectral
Analysis and its Applications (Holden-Day,
San Francisco, 1968).
[11] Chatfield, C. , The Analysis of Time Series:
Theory and Practice, Chap. 4 (Chapman and
Hall, London, 19/b).
[12] Bateman, D. N., Leeman, S., Metreweli, C,
and Willson, K. , A noninvasive technique
for gastric motility measurement, Brit. J.
Radiol. 50, 526-527 (1977).
[13] Leeman, S., Badcock, P. C, Gore, J. C,
Plessner, N. J., and Willson, K., Ultra-
sonic Backscattering Assessment of Tumour
Response to Treatment, presented to Tumour
Ultrasound 77, International Conference,
London, 1977. (Copy available by request.)
280
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer , ed.. National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
COMPUTER SPECTRAL ANALYSIS OF ULTRASONIC A MODE ECHOES
D. E. Robinson
Ultrasonics Institute
Sydney, Australia
Tissue attenuation as a function of frequency measured by the effect on the power
density spectrum of the echo off glass shadowed by the tissue sample is shown to give
results comparable to those published previously. A method of measuring attenuation
within tissue by comparison of the scattered echoes from shallow and deep scatterers
is investigated and shown to have limitations. An assessment of properties of the
scatterers from investigation of the spectral properties of the echoes from a scatter-
ing region is suggested.
Key words: Acoustic; computer processing; digital acquisition; digital signal
processing; pulse-echo techniques; spectrum analysis; ultrasonics.
1. Introduction
At present the bulk of tissue characterisation
by ultrasound in clinical practice is done by
qualitative examination of the grey scale B Mode
echogram. The current work is directed to deriv-
ing quantitative characterisation data by spec-
tral analysis of 180° backscattered echoes, or
ultrasonic A Mode, following the approach of
Lizzi [l]i.
2. Measuring Equipment
The transducer beam pattern requires to be nar-
row to give good spatial resolution and with a
well behaved spectrum to avoid complications in
spectral analysis techniques [2]. This is achiev-
ed in the focal zone of a focussed transducer of
aperture 65 mm and curvature 35 cm and usable fre-
quency range from 1.5 to 4 MHz. The 20 dB echo
beamwidth at the focus is 6.7 mm.
The receiver amplifier gain is variable in 1 dB
steps allowing an input dynamic range of 100 dB
without overload. Receiver gain variation is used
instead of transmitter power control as it is found
that due to non-linearities in the transmitter, the
transducer or the medium, significant changes in
echo wave shape occur with changes in transmitter
power. The signals are digitised at 10 MHz sam-
pling rate using an 8 bit Biomation 8100 waveform
recorder interfaced to an Interdata Model 85 com-
puter. The computer and an interactive signal
processing program have been described elsewhere
[3].
The spectrum used here is the square magnitude
of the Discrete Fourier Transform (DFT) of the
echo signal or Power Density Spectrum. All spectra
shown in this paper have been interpolated as de-
scribed by Rabiner et al. [4] by extending the set
^Figures in brackets indicate literature
references at the end of this paper.
of 128 or 256 recorded time samples with zero-
valued samples up to a total signal length of 1024
before computing the DFT. Comparisons between
spectra are made by calculating the logarithm of
the ratio of the spectra (log ratio) and observing
the ordinate at 2.5 MHz in dB and the slope in
dB/MHz. The log ratio for point and plane target
echoes remain within 3 dB and the slope within
2 dB/MHz for distances from the transducer focus
of ± 30 mm.
A B-Mode display of the examined tissue is pro-
vided for A mode beam position selection and guid-
ance. An example of input data and corresponding
B mode display is shown in figure 1.
■ 3. Experimental Techniques
First the echo from a thick glass block at the
focus with no intervening tissue is recorded, and
regarded as the impulse response of the system.
The tissue specimen is then placed in a poly-
ethylene bag in front of the glass block and echoes
from a series of 30 adjacent beam positions at 3 mm
intervals were recorded from three locations; from
the glass block behind the tissue and from shallow
and deep within the examined tissue. In each case
the transducer was moved so that the echoes were
recorded from the focal area of the transducer
beam. In the case of the scattered echoes, a
region was selected which did not have large dis-
crete echoes on the B Mode scan.
4. Signal Processing
When measuring the spectrum of a statistical sig-
nal, some form of averaging is required to obtain a
smooth and stable spectral estimate [5]. This can
be done in the spatial domain by translating the beam
axis as described by Lizzi [1] and adding the spectra
corresponding to a number of adjacent lines of sight
to obtain an average spectrum. This has the effect
of reducing the amplitude of ripples in the spectrum.
281
Fig. 1. B-Mode echogram showing tissue sample
in front of glass block. The two
lines indicate the area used for A
Mode acquisition. The complete A mode
signal and the part of the signal near
the focal region are also shown.
It should be noted that the "fine structure" in the
spectrum of individual echo signals may be a func-
tion, not of the statistical nature of the sample,
nor of the local tissue structure, but simply of the
length of the time window. For instance, given a
signal x(t) with transform X(f) and spectrum X(F)2
the sum of two identical shifted signals x(t) +
x(t - t) has a spectrum 2X(f)2 (1 + cos wt). The
factor (1 + cos wt) gives rise to "wiggles" in the
measured spectrum which are not directly related to
local tissue "structure" but are merely phase ef-
fects between widely spaced echoes. Figure 2a and
2b show the spectra for a single echo and two identi-
cal echoes separated by 5 mm. This effect is re-
duced by windowing in the autocorrelation domain, or
lag windowing. The autocorrelation function is the
Inverse Fourier Transform of the Power Density Spec-
trum. The Autocorrelation functions of the single
echo and two identical echoes spaced at 5 mm (66
sample intervals) are shown in figure 2c and 2d.
Also shown in figure 2d is a Hamming window in the
Autocorrelation (lag) domain suitable for removing
the effect of widely spaced echoes. Multiplication
by the window function has the effect of removing
contributions to the Autocorrelation function of
lags greater than 32 sample intervals (2.5 mm) and
leaving only the contributions from closely spaced
echoes. The smoothed Power Density Spectrum ob-
tained by re-transforming the weighted Autocorrela-
tion function into the frequency domain is shown in
figure 2e. The effect of this procedure on tissue
echoes is shown in figure 3 where the original time
sample of scattered echo data is windowed with a
Hamming window of length 128 (corresponding to 10 mm
of tissue). The lag windows are of length 64, 32
and 16 samples increments corresponding to correla-
tion lengths of 5, 2.5 and 1.25 mm and thus limit-
ing the effect on the spectrum to structures of
these sizes.
In the analysis of the echoes off glass no time
windowing or spectral smoothing is necessary since
we are dealing with only deterministic single echo
waveforms and not statistical signals. The power
density spectra of the echo off glass with and with-
out shadowing by the tissue were calculated, and the
log ratio formed. A typical set of results is shown
in figure 4. The log ratio is a measure of the at-
tenuation due to the tissue as a function of fre-
quency. The values are only significant for fre-
quencies contained in the original power density
spectrum of the unshadowed glass echo. A line of
best fit was computed on a least total weighted
squared error basis with the weighting function be-
ing the original power density spectrum values for
the echo off glass at the focus.
The parameters of the line y = ax + b where y is
the best fit to the log ratio value and x the fre-
quency value are given by:
a =
? W. ? x. y. W. - ? x. W. ? y. W.
1 11 1-^11 1 11111
and b =
? W. ^ x? W.
? W. ^ x. W.
? w.
1 1
? w.
1 1
282
Fig. 2. a) Power density spectrum of a single
echo off glass; b) spectrum of two
echoes separated by .5 mm; c) auto-
correlation function corresponding to
spectrum in a; d) autocorrelation
function of spectrum shown in b and
64 point Hamming lag window;
e) smoothed spectrum obtained from
the lag windowed autocorrelation
function shown in d.
where the x., y. and W. are the values of the fre-
quency, logVatIo and vveighting function for the
successive frequency samples.
For echoes from scatterers within the tissue
the echo signal was windowed using a time window
of length 128 samples or 10 mm and the spectrum
was smoothed by a lag window of 64 lags correspond-
ing to 5 mm. The log ratio between deep and shal-
low echo spectra was formed and the weighted best
fit line calculated as before. A typical result is
shown in figure 5.
5. Results
Being initial experiments readily available ani-
mal material was used. The material was skeletal
beef muscle along and across the fibre direction
and calf liver. The results are shown in table 1.
The log ratio curve, indicating tissue attenua-
tion was approximately linear within the pass band
for all tissue samples using the shadowed glass ap-
proach. For this reason more confidence is placed
in the determination of attenuation slope by this
283
Fig. 3. Effect of lag windowing on echo spectrum. A single line of data was windowed
with a 1 cm Hamming window and the spectrum a) smoothed with a Hamming lag
window of b) 5 mm, c) 2.5 mm and d) 1.25 mm.
Table 1. Tissue attenuation determined by the shadowed glass
method and by the scattered echo comparison method.
Tissue Quoted atten. Shadowed glass Scattered
(Wells [6]) Slope Atten. at 2.5 MHz Slope Atten. at 2.5 MHz
dB/cm MHz dB/cm MHz dB/cm dB/cm MHz dB/cm
Muscle along
f i bres
1.3
1.5
4 .9
1.2
5.8
Muscle across
f i bres
3.3
(1)
.65
1.5
.28
2.8
(2)
.78
1.8
.15
1.7
Li ver
.94
(1)
.82
2.2
.72
3.0
(2)
.73
3.1
.53
4.2
284
Fig. 4. Results of shadowed glass measurement on muscle across fibre direction showing
a) unshadowed and b) shadowed spectra, c) ratio of spectra and d) log ratio
with weighted best fit line superimposed. Note that the ripple amplitude is
small in the linear region.
method. The values attenuation versus frequency
in dB/cm MHz were in reasonable agreement with pre-
viously published values for the muscle along the
fibres and for liver. The values for muscle across
the fibres were similar to the other values but not
in agreement with Wells [6]. The log ratio curves
for the scattered echo measurements all showed
significant ripples, indicating that some phasing
effects still remain. The values for liver and
muscle along the fibres agreed reasonably well with
the shadowed glass method, but the muscle along
fibres gave widely different results.
6. Discussion
It is obvious that further work is needed before
the scattered echo comparison method will be useful
for attenuation determination. The shadowed glass
method appears to give repeatable results but is not
suitable for clinical examinations.
The current approach in scattered echo analysis
is to average in the power spectral domain with the
aim of reducing the ripples in the spectrum to ob-
tain a smooth and stable spectral estimate. It ap-
pears that an alternative approach is to average
in another domain to obtain a stable estimate of
the underlying spectral ripple, which is a measure
of the "structure" of the echo signal. Two obvious
candidates are the Autocorrelation and the cepstral
domains. This structure can be postulated from the
same model of distributed discrete point scatterers
that is used to justify the single frequency scat-
tering versus angle approach to tissue characteri-
sation .
References
[1] Lizzi, F., Katz, L. , St. Louis, L., and Cole-
man, D. J., Applications of spectral analysis
in medical ultrasonography. Ultrasonics 14, 77-
80 (1976).
[2] Robinson, D. E. and Williams, B. G., Computer
Analysis of Ultrasonic Pulse Echo Signals, in
Ultrasound in Medicine and Biology, D. White
and R. E. Brown, eds., p. 1443 (Plenum Press,
New York 1977).
285
6db
j M 1
*
m
Fig. 5. Results of scattered measurement showing a) shallow and b) deep echo spectra,
c) the ratio and d) log ratio with best fit line superimposed. Note the
large ripple content of the log ratio.
1 n
[3] Robinson, D. E., Williams, B. G., and Horn,
P. R., Digital acquisition and interactive
processing of ultrasonic echoes. Ultrasound
Med. S Biol . 2, 199-212 (1976).
[4] Rabiner, L. R. , Gold, B., and McGonegal, C. A.
An Approach to the Approximation Problem for
Nonrecursive Digital Filters, in IEEE Trans.
Audio Electroacoustics , AV-IS, 83-106 ( 1970).
[5]
[6]
Jenkins, G. M. and Watts, D. G., Spectral
Analysis and its Application, p. 209 (Holder-
Day 1968).
Wells, P. N. T. ,
sonic Diagnosis.
Physical Principles of Ultra-
Academy Press 1969.
286
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed., National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
CEPSTRAL SIGNAL PROCESSING FOR TISSUE SIGNATURE ANALYSIS
J. Eraser and G. S. Kino
Stanford University
Stanford, California 94305, U.S.A.
J. Birnholz
Harvard Medical School
Boston, Massachusetts 02115, U.S.A.
The reflected signal received by an ultrasonic transducer is modeled as a convolu-
tion of a transducer response with a reflection function for the target region.
Cepstral analysis translates that signal into a domain where those components inter-
act additively rather than through convolution and where separation can be accomplish-
ed with simple bandpass filtering operations. The cepstral transform of the pulse
echo signal also provides direct access to any periodic behavior Ci' reflectors aris-
ing from their spacing. As an example of these capabilities, the technique is used
experimentally to describe the characteristic spacing of macrostructural reflecting
elements in the pig liver in vitro and to determine the frequency dependent attenua-
tion behavior of normal human liver in vivo.
Keywords: Attenuation; B-scan; cepstrum; computer; convolution; de-convolution;
liver; power spectrum; signal processing; tissue characterization;
tissue parameters; ultrasound.
1. Introduction
We describe in this paper preliminary experi-
ments with new types of non-linear signal proces-
sing used for eliminating certain types of signal
distortion in conventional acoustic imaging sys-
tems which preclude unambiguous identification of
specific tissue features.
An acoustic propagation path through a tissue
region can be treated as a one-dimensional array
of reflectors. While each individual reflector
may be expected to replicate the time charac-
teristics of a self convolved transducer impulse
response (assuming an impulsive excitation), that
pulse form is altered or distorted during propaga-
tion by the frequency dependent attenuation
processes of the medium intervening between re-
flector and transducer. Previous studies have in-
dicated that tissue specific information is de-
rived from both attenuation properties of the
region, as well as the distribution and backscat-
ter features of the individual reflectors [1]^.
Therefore, an effective signal processing scheme
must provide a means of eliminating the trans-
ducer response and separating frequency dependent
attenuation effects from target signal components.
Our approach has been to take advantage of certain
properties of the cepstrum, or Fourier transform,
of the signal log power spectrum [2]. The ration-
ale for this choice derives from previous uses in
^Figures in brackets indicate literature
references at the end of this paper.
comparable signal processing tasks: (1) homo-
morphic deconvol ution of speech and music wave-
forms for improved recording fidelity [3]; (2)
dereverberation enhancement of seismic data [4];
and (3) recognition and characterization of flaw
configuration in nondestructive testing [5].
This method has not been applied to medical diag-
nostic ultrasound previously.
2. Analysis of the Return Echoes
We consider first what occurs when a signal is
incident on body tissue. We will mainly be in-
terested in the situation in which we observe a
reflected signal with the same transducer used to
iTluminate the tissue. Techniques for looking at
off axis reflections can, in principle, provide a
great deal more information. But, in practice,
they are extremely difficult to employ without the
use of arrays of transducers to carry out repeat-
able measurements of this kind. Even with the use
of arrays of transducers, because of the distor-
tions along the body path to an organ of interest,
the problems are very severe in terms of repeat-
ability. In that case, it is probably better to
concentrate on the development of imaging systems
first, then process the images in some detail.
Thus, the method described here is entirely devot-
ed to the use of a single transducer operating in
a reflection mode.
The signal received by a reflection mode system
can be considered to arise from a one-dimensional
reflection function R of distance z along the beam
path R(z). In the far field of a plane transducer.
287
or the focal zone of a focused transducer, the
phase fronts are roughly planar and normal to the
beam. Under these conditions, it can be shown by
perturbation theory that R(z) is the weighted sum
across the beam of the component along the beam
of the gradient of the acoustic impedance of the
tissue. That is, if the beam pattern is P(r), and
the impedance is Z(r,z) which varies slightly
around an average value of Zg,
R(z) - y 2TTrdrP{r)z • v|z(r,z)f . (D
° 0
We wish to work in the time domain, using R(t),
where, with V the acoustic velocity in tissue
t = 2z/V (2)
since the time required is for a round trip. Any
information about R(t) which can be obtained will
be related to changes of impedance in the tissue,
and, hence, may be of use in tissue characteriza-
tion.
A region of tissue is interrogated by passing
a pulse A(t)of ultrasound through it and observ-
ing the return signal S(t). S(t) is given by the
convolution of A(t) and R(t):
S(t) = /A(t) R(t - T)dt (3)
which can be represented symbolically as
S{t) = A(t) * R(t). (4)
We do not necessarily seek to recover R(t) exact-
ly, but rather to derive statistical information
about it. We are interested in knowing the distri-
bution sizes of the objects response for the re-
flected signal, and how regularly they are spaced
in the tissue. We may also be interested in the
evolution of A(t) as the pulse passes through tis-
sue with frequency dependent attenuation.
Since the autocorrelation function responds to
structures with strong peaks, it could be con-
sidered a possible processing technique to extract
structure information from S(t). However, it can
be shown that the autocorrelation of two convolved
functions is the convolution of their autocorrela-
tions. Using -A" to represent correlation,
S(t)*S(t) = (A(t)*A(t)) * (R(t)*R(t)) . (5)
Since A(t) is an oscillatory function, A(t)*A(t)
is also, and the autocorrelation of R is confused
by the convolution. The best that can be done in
this case is to recover the envelope of the auto-
correlation. A convenient technique for this pur-
pose is to find the envelope of the analytic func-
tion associated with the autocorrelation by taking
the Fourier transform, zeroing negative frequency
components, and taking the magnitude of the inverse
transform [6]. Since the autocorrelation is normal-
ly performed by inverse transforming the power
spectrum, very little additional effort is needed.
One merely zeros the negative frequencies in the
power spectrum.
An example of a typical ultrasound echo is
shown in figure 1. This echo represents the re-
turn from a pig liver, using a 3.5 MHz transducer.
A pattern of multiple echoes is evident, but not
readily quantifiable. The envelope of the auto-
TIME , MICROSECONDS
Fig. 1. The recorded ultrasonic echo from a
section of fresh pig liver, in vitro,
using a nominally 3.5 MHz transducer.
This represents 1.9 cm of tissue.
0 12.8
TIME , MICROSECONDS
Fig. 2. The envelope of the autocorrelation
function of the signal in figure 1.
correlation of this signal is shown in figure 2.
Several peaks are discernable, but none is dominant.
A more sophisticated technique is sought.
3. The Inversion Process
In tissue signature analysis, responses from
the tissue are weighted by the form of the trans-
ducer response. In addition, there is distortion
due to the tissues on the way to the organ of
interest. So a process is desired which, given
S(t), can separate A(t) and R(t). Under ideal
conditions, if one of two convolved functions is
known, the other may be found. Deconvolution is
performed by Fourier transforming, inverse filter-
ing, and inverse Fourier transforming.
If eq. (4) is Fourier transformed, the convolu-
tion leads to a product, and rearranging gives:
RM = iM. . (6)
Several factors prevent the use of this formal in-
version process in the present case. The presence
of finite noise in the recorded signal means that
S(oj) is not known exactly. In addition, as the
ultrasonic transducer responses are band-limited,
A(u) becomes vanishingly small in the same frequen-
cy ranges where S(a)) is dominated by noise, so the
noise has a disproportionate influence on the re-
covery of R(a)). Finally, since the signal has prop
propagated through a tissue of unknown and frequen-
cy dependent attenuation, A(a)) is not known. A
different approach is, therefore needed.
288
4. Homomorphic Processing
We have used so-called homomorphic processing,
or cepstral analysis, for this purpose. At the
present time, we have by no means developed this
type of processing to its limit, but we have made
a start on techniques employing it and our inten-
tion is to develop the techniques further.
The basic idea behind homomorphic processing
is to turn a nonlinear process into a linear
process.
Consider the convolution to the signal from
the transducer A(t) with the tissue response
R(t) in the time domain
S(t) = A(t) * R(t)
(7)
In the frequency domain, the two components de-
rived from these signals are
A(a)) • R(a))
(8)
A product form is also obtained for the power
spectrum
tS(a,)|2 = |A(a.)|2 . |R{^)|2 (9)
By taking the logarithm of |S(q))|2, the multipli-
cation is converted to a sum. This enables
normal linear signal processing techniques to be
used.
Let
then
S(a>) = log|S(a))|2
A{o)) = log|A(a>)|2
R(a)) = log|R(a))|2 ,
S(a)) = A(a)) + R(oj) .
(10)
(11)
(12)
(13)
Finally, if we take the Fourier transform of
S(a)) defined as S(t), we can write
S(t) = A(t) + R(t)
(14)
S(t) is called the cepstrum of S(t). The nota-
tion is used to distinguish it from S(t).
Figures 3, 4, and 5 demonstrate the convolution
of a simulated transducer impulse response with
an impulse pair, and the additive property of
the log power spectra and cepstra.
We note that in the log power spectrum domain
the system is linear and subject to all the rules
and techniques of linear filtering. Such filter-
ing may be accomplished by taking one more
Fourier transform, yielding the cepstrum of the
LjJ
Q
Z)
CL
<
CO
(a)
(b)
(c)
TIME
Fig. 3. (a) A simulated transducer impulse
response of the form
g-^(t/8)2 2TT(t/8).
(b) An impulse pair with spacing 16.
(c) The convolution of (a) and (b).
A
A
FREQUENCY
Fig. 4. The log power spectra of figure 2(a),
(b), and (c) respectively.
289
(a)
(b)
(c)
TIME
Fig. 5. The cepstra of figure 3(a), (b), and
(c) respectively.
ceps'trum of the reflection pattern is not known
and varies from tissue to tissue.
Some examples of the cepstra of impulse trains
can yield insight into what may happen in tissue,
and suggest features which will be meaningful.
Two im.pulses with separation t will yield a string
of impulses with separation t and of strength de-
caying fairly rapidly toward zero away from the
origin, as seen in figures 3(b) and 5(b). If more
than two impulses exist with the same spaces, the
cepstrum is similar, but with a slower decay rate.
This is depicted in figure 6. Figure 6 also de-
monstrates a problem which often arises in the
digital implementation of the cepstrum. When the
logarithm of the power spectrum is taken, the non-
linear process generates harmonics in the time
domain which can cause aliasing in the cepstrum,
even though the original spectrum bandwidth was
well within the Nyquist bandwidth of the sampling
process. In general, any repetitive pattern in a
signal will lead to a peak at a time equal to that
spacing in the spectrum, as well as lesser peaks
at multiples of that spacing. The cepstrum itself
may not be easy to interpret for ultrasound echoes
from tissue unless the tissue has a fairly simple
and distinct structure. But the decay rate of the
cepstrum is related to the spacings and numbers of
periodic reflections in the tissue, and hence, may
be useful for tissue characterization.
If the minimum spacing between strong reflec-
tions in the tissue is somewhat greater than the
wavelength of the interrogating sound, all informa-
tion about the periodicities of reflections will
be separated from the information about the charac-
ter of an individual reflection in the cepstrum.
In this case, it should be possible to filter the
cepstrum and recover the average of the power
spectral densities of the individual reflections
from a region of tissue. This should be a good
original signal. Filtering may then be perform-
ed by multiplying the cepstrum by a weighting
function to select a desired portion, then re-
versing as many processing steps as necessary to
reach a useful and recognizable domain. Since
information was lost in taking the power spec-
trum complete inversion is not possible. This
restriction is removed by the use of the complex
cepstrum, in which the real logarithm of the
power spectrum is replaced by the complex loga-
rithm of the complex spectrum. However, this
technique, which has been demonstrated in seis-
mographic and speech processing applications, is
much more difficult to use in practice. If, in-
stead, we reconstruct the signals from the real
cepstrum, we obtain the correlation function of
the original signals, or of the tissue impedance
variations.
5. Possible Applications
Filtering in the cepstrum domain is useful only
if the cepstra of the two components fall in dif-
ferent regions of the domain. The spectrum of a
broadband transducer impulse response can be close
to a Gaussian response in frequency. So its loga-
rithm is close to a parabolic response in fre-
quency, and cepstrum is a relatively narrow pulse
centered at the origin, which has a time width
of the same order as the original impulse. This
effect can be seen in figures 3(a) and 5(a). The
(a)
(b)
TIME
Fig. 6.
(a) Four impulses with spacing 16.
(b) The cepstrum of the convolution
of the impulse response of 3(a),
with the impulse train of 6(a).
290
approximation to the spectrum of the transmitted
ultrasound pulse as it passes through the region
of tissue within an unknown constant. If such a
calculation is carried out on signals from two
succeeding regions of tissue along the line of
sight, the difference between them will contain
information about the frequency dependence of the
attenuation of ultrasound in the tissue between
these two regions.
6. Experimental Techniques
A surplus B-scan imaging system, loaned by the
Radiology Department of the Stanford University
Medical Center, was obtained, originally as a
means of positioning the ultrasound transducers
and generating images for identification of the
source of echoes to be processed. It was found
that the pulse generator and receiving amplifier
of the system, a Picker, were of excellent quali-
ty. Also, a depth marker was available, which
could be used to select a portion of a trace.
By tapping a small amount of the received signal
with a high impedance isolation transformer at a
low impedance point in the receiver, a high quali-
ty radio frequency signal was obtained without
compromising either the operation or safety of the
existing B-scan system. The tapped R-F signal was
further amplified and monitored by an oscilloscope
and a Biomation 8100 transient recorder, which
was set to record when armed by a pushbutton ac-
cessible to the operator of the B-scan system.
A PDPll/10 minicomputer was used to store the
digitized R-F signals from the transient recorder
on a magnetic disc, and later to read the stored
records and process them as desired. In practice,
the operator manipulated the transducer until
the beam passed through the desired area, using
the B-scan image, if desired; then placed the
depth marker at the beginning of the region of
interest, checked the signal on the oscilloscope,
and initiated recording with the pushbutton.
This was found to be quite fast and convenient.
The system is depicted schematically in figure 7.
After some experimentation, it was decided to
standardize the format of recorded data in order
to simplify the signal processing programs.
A sample rate of 20 MHz was chosen, giving a
B-SCAN
SYSTEM
TRONSOUCER
PUL5ER
— r~
DATA ACQUISITION PROCESSING
a
STORAGE
OSCILLOSCOPE
TRANSIENT
RECORDER
n -J
MANUAL
ARMING
CONTROL
[— MINICOMPUTER
MAGNETIC
DISC
ANALOG SIGNAL PATH
DIGITAL SIGNAL PATH
Fig. 7. A schematic diagram of the system used
to obtain, store, and process ultrasonic
echoes .
usable signal bandwidth of 10 MHz. Record
lenngth was standardized at 512 samples, which
represented 25.6 ys, or about 1.9 cm of tissue.
This was found to be long enough for most pur-
poses, but a provision was made to store four
consecutive records if desired, representing
7.7 cm of tissue. Fourier transforms could be
calculated on 512 point records in 10 seconds,
and cepstra in 23 seconds, fast enough for inter-
active studies on the effects of various types
of filtering.
Two types of experiments have been run to col-
lect data for various purposes. In the first
type, the scanning arm of the B-scan system is not
used. The transducer is fixed in a water tank,
suspended from a micrometer adjustable frame.
The transducer's beam could be aligned to reflect
from the water surface for transducer characteriza-
tion and reference purposes, on targets suspended
from a rotatable mount, or on tissue samples pin-
ned to a sponge on the bottom of the tank. In the
second mode, the scanning arm was used and identi-
fication of tissues was made using the images
formed.
Single reflections from the surface of the
water tank were used to obtain data on the effi-
ciency and spectral characteristics of both pur-
chased and fabricated transducers, and to verify
the structure of the cepstrum of a transducer im-
pulse response. Impulse responses and cepstra for
two commercial transducers are shown in figures 8
and 9. A transducer was then aimed at the rotat-
ing mount, and a fixture holding three 1.1 mm
diameter rods side by side in the beam at a spac-
ing of 2.5 mm was attached, as shown in figure 10.
The array of rods was rotated from perpendicular
to the beam to parallel, and on to perpendicular
again, and the reflections recorded at five degree
increments in angle to provide reflections of
various spacings. The envelope of these cepstra
is plotted as a function of angle in figure 11.
The sinusoidal dependence of spacing on angle can
be clearly seen. The short spacing is seen best
on one side of 90°, and the long spacing on the
(a)
(b)
25.6
TIME . MICROSECONDS
Fig. 8.
(a) Impulse response of a commercial
2.25 MHz transducer.
(b) Impulse response of a commercial
3.5 MHz transducer.
291
12.8
TIME , MICROSECONDS
Fig. 9. (a), (b) Cepstra of the impulse responses
of 8(a) and (b) respectively.
3.2 6.4 9.6
TIME, MICROSECONDS
ANGLE B
Fig. 11. Moduli of cepstra recovered from the
scheme of figure 10, as a function of
angle.
ROTATING
MOUNT
CYLINDRICAL
RODS
TRANSDUCER
Fig. 10. Schematic diagram of a scheme to produce
three identical echoes with variable spac-
ing. Three 1.1 mm diameter steel rods are
mounted 2.5 mm apart center to center.
other because a slight misalignment of the rods
caused the reflections to be unequal in strength
and to depend on angle. At angles where the
separation between the reflections is comparable
to the pulse width (that is, about .5 ys), inter-
ference can be seen to occur between the peak rep-
resenting the pulse shape and that representing
the periodicity.
In order to demonstrate the ability of the cep-
strum to detect a periodic structure in biological
tissue, a particularly simple sample was chosen:
pig liver. Pig liver is a large organ, readily
available in a fairly fresh state, and having a
characteristic and simple period structure [8].
Our sample was several centimeters thick, and was
observed to have an internal structure consisting
largely of spherical lobules of fairly uniform
size, about 2 mm in diameter. Several echoes were
recorded from various parts of the liver, using a
3.5 MHz focused transducer in the water tank. The
best of the cepstra recovered is shown in figure
TIME , MICROSECONDS
Fig,
12. The modulus of the cepstrum of the signal
shown in figure 1 .
12. The signal from which it was derived was
shown in figure 1. The strong peak at 2.5 ys cor-
responds to a spacing of 1.9 mm between reflectors,
which corresponds well with the observed size of
the lobules. Other cepstra were not always so
simple.
Cepstra were also calculated for signals from
various regions of human livers from autopsies,
as well as from thigh muscles and the area of the
liver, in vivo. The details of the structure of
both the power spectrum and the cepstrum were
found to vary markedly among the samples, and
even between two consecutive 5 mm long regions of
single records. Some cepstra showed substantial
single peaks; some did not. The decay rates of
the cepstra varied substantially also. We conclude
that the tissues studied are sufficiently inhomo-
geneous to prevent easy interpretation of these
results. A broad statistical study of a large num-
ber of signals documented as to the location and
possible pathology of the tissue, original trans-
ducer response, and thickness of intervening tis-
sue, will be needed to prove the usefulness, or
lack thereof, of such techniques.
292
7. Attenuation Estimation
Since the cepstrum is the Fourier transform of
the log power spectrum, low pass filtering, or
smoothing, of the log power spectrum may be done
by multiplying the cepstrum by a window function
which leaves short time responses untouched and
eliminates long time values. The resultant gated
signal may then be Fourier transformed to the
frequency domain again. If the variations in the
spectrum are due to reflections with relatively
large spacing tissue, and, hence, are rapid com-
pared to the finite bandwidth of the incident beam
of sound, the rapidly varying spectral components
can be filtered out, leaving a good approximation
to the log power spectrum of the incident beam
distorted by the average attenuation in the tis-
sue. If such an approximation is made for two
consecutive regions of tissue along the same
ultrasonic beam path, the difference between them
should exhibit the frequency dependence of the at-
tenuation of ultrasound through the path between
them. Such a calculation does not require as
strong an assumption about the variations in the
log power spectrum due to the tissue reflections.
There may be slow variations; if they are the same
in the two regions of tissue, their difference
will be zero. Since slow variations in the log
power spectrum arise from structure fine compared
to the ultrasonic pulse length, the success or
failure of this technique to predict a plausible
frequency dependence for the attenuation of ultra-
sound will give a measure of the similarity of the
structure of the two regions of tissue on the
scale of the ultrasonic pulse length.
Since the advantage of this technique would be
its ability to measure attenuation in vivo non-
invasively, it was applied to the data recorded
-+20r
UJ
o
cl
CD
o
(b)
■201-
FREQUENCY (MHz)
Fig. 13. (a) and (b), log power spectra from two
consecutive 1 cm long regions of in vivo
human liver tissue with 2 cm center spac-
ing, recorded with a 2.25 MHz transducer.
1" This procedure is illustrated in figures
13 and 14. First the data is windowed to the pro-
per length and Fourier transformed, and log power
and spectra obtained. The signal to noise ratio
has been artifically decreased to 20 dB to de-em-
phasize the random spectral components outside the
useful bandwidth of the transducer, 1 to 3 MHz.
The cepstra are then computed and low pass filtered
with a sine-squared window, which closely approxi-
mates a Gaussian window, but completely cuts off
long time components. Fourier transforming again
yields the smoothed log power spectrum. Subtract-
ing one such spectrum from the other and dividing
by the path length between them yields a curve
which, over the useful bandwidth of the transducer,
estimates the frequency dependence of the loss over
that path. The absolute loss is not estimated here,
although with attention to time-gain control
settings and assumption of equal reflection ampli-
tudes in the two regions, it could be. In this
example in which the signals were taken from two
regions about 1 cm long and 2 cm apart in the area
of the liver of a person with a presumed normal
liver, the estimated loss follows fairly well a
linear frequency dependence, although with a coef-
ficient of 0.8 dB/cm MHz, slightly lower than the
previously reported in in vitro value [9]
20r
UJ
o
Q.
O
T3
<
ZD
111
>
<
FREQUENCY (MHz)
Fig. 14. (a) Solid line, the cepstrally smoothed
log power spectrum derived from that
of figure 13(a). Dashed line, the
same function of the spectrum of
figure 13(b).
(b) Solid line, the difference between
the two smoothed spectra, divided
by the path length. Dashed line,
the slope of the attenuation in the
frequency range where both spectra
are above noise. This slope is
.8 dB/cm.
293
-+ 20
LU
20
0
•20r
(a)
FREQUENCY (MHz)
Fig. 15. Log power spectra analogous to those of
figure 13, but from a different region
of the liver.
An interesting phenomenon has been found in
the two vn_ vivo livers and the one i_n_ vitro liver
studied with this technique so far. In many of
the log power spectra calculated, there is a deep
notch, representing an absence of reflection at
frequencies varying from 2.4 to 2.7 MHz. This
notch has been seen with both focused and unfo-
cused transducers of both 2.25 and 3.5 MHz center
frequency. This phenomenon violates the condi-
tions under which one can get a good estimate of
attenuation. It occurs only once in the useful
bandwidth of the transducers used, and so cannot
be considered rapidly varying. Its frequency
varies enough from region to region that it does
not cancel when two spectra are subtracted. The
very deep notch is almost, but not quite, re-
moved by the smoothing operation. Since the
attenuation calculation takes the difference be-
tween two similar functions, it is sensitive to
the small error which remains. An example of this
problem is shown in figures 15 and 16. Such a
phenomenon probably arises because most of the
reflections in a region are coming from a single
type of structure: one with a constant size of
a fraction of a millimeter. Measurements over a
broader bandwidth could determine more exactly
what type of structure this is. If it could be
identified, there might be clinically useful in-
formation available.
8. Conclusions
From this limited study, we conclude that:
(a) Cepstral analysis, by converting ultra-
sonic signals from a domain in which the output is
the convolution of an impulse response with a re-
flection function to a domain in which it is a
sum, allows the separation of information about
the distribution of spacings between reflectors
from information about the incident impulse.
LU
B
a.
o
o
20
-
/ \
\ \
\ /
\ ^
\
1 1 I
(a)
o
<
UJ
>
UJ
cr
FREQUENCY (MHz)
Fig. 16. (a) Smoothed log power spectra from those
of figure 15.
(b) Attenuation estimated from figure
16(a). No estimate is possible in
this case.
(b) Two potentially useful parameters for tis-
sue characterization in vivo are the form of the
cepstrum of ultrasound echoes from tissue, which
can identify periodicities in the tissue; and at-
tenuation coefficient estimation by cepstral
smoothing of log power spectra.
A lack of consistency from one region to
another within a tissue suggests considerable
variability in the tissue. Further work, utiliz-
ing statistical studies on large data bases, will
be required to verify the usefulness of these
techniques. Cepstral analysis should prove use-
ful, since averaging over similar samples will en-
hance the tissue-dependent information. This is
in marked contrast to the power spectral domain,
where averaging over similar samples tends to
eliminate tissue-dependent information and approxi-
mate the power spectrum of the measuring system.
Acknowledgments
The work reported in this paper was supported
in part by the National Science Foundation under
Grant No. ENG 75-18681, and in part by the Kaiser
Foundation.
References
[1] Several of these studies are reviewed in:
Dunn, F., Ultrasonics Attenuation, Absorption,
and Velocity in Tissues and Organs; and,
Reid, J.M., The Scattering of Ultrasound by
Tissues, in Ultrasonic Tissue Characterization.
M. Linzer, ed.. National Bureau of Standards
Spec. Publ. 453, pp. 21-28 and 29-47 (U.S.
Government Printing Office, Washington, D.C.,
1976).
294
[2] Bogert, B. P., Healy, M. J. R., and Tukey,
J. W., The Frequency Analysis of Time Series
for Echoes: Cepstrum, Pseudo-Autocovariance,
Cross-Cepstrum, and Saphe Cracking, in Proc .
Sym. Time Series Analysis, M. Rosenblatt, ed. ,
pp. 209-243 (John Wiley and Sons, Inc., New
York, 1963).
[3] Stockham, T. C, Jr., Restoration of Old
Acoustic Recordings by Means of Digital Signal
Processing, Preprint, 41st Convention, Audio
Engineering Society, New York, October 1971.
[4] Ulrych, T. J., Application of homomorphic
deconvolution to seismology. Geophysics 36
(4), 650-660 (August 1971).
[5] Adaptive Nonlinear Signal Processing for
Characterization of Ultrasonic NDE Waveforms,
Task 2: Measurement of Subsurface Fatigue
Crack Size, in Technical Report AFML-TR-76-44,
April 1976, Section 6.4, p. 54.
[6] Bracewell , R., The Fourier Transform and Its
Application, pp. 269-27MMcGraw-Hi 1 1 , New
York, 1965).
[7] Oppenheim, A. V. and Schafer, R. W., Digital
Signal Processing Chap. 10, pp. 480-531
TPrentice-Hall , Inc., Princeton, 1975).
[8] Lele, P. P., Mansfield, A. B., Murphy, A. I.,
Namery, J., and Senapati , N., Tissue Charac-
terization by Ultrasonic Frequency Dependent
Attenuation and Scattering, in Ultrasonic
Tissue Characterization, M. Linzer, ed..
National Bureau of Standards Spec. Publ. 453,
pp. 167-196 (U.S. Government Printing Office,
Washington, D.C., 1976).
[9] Goldman, D. E. and Heuter, T. F., Tabular
data of the velocity and absorption of high
frequency sound in mammalian tissues, J_.
Acoust. Soc. Am. 28, 35 (1956).
295
1
t
li
ij
If
is
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed. , National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
RECOGNITION OF PATTERNS IN ULTRASONIC SECTIONAL PICTURES
OF THE PROSTATE FOR TUMOR DIAGNOSIS
W. von Seelen,^ A. Gaca,^ E. Loch,^ W. Scheiding,^ and G. Wessels^
ilnstitut f(Jr Zoologie
Arbei tsgruppe III (Biophysik)
D-6500 Mainz, Federal Republic of Germany
^Deutsche Klinik fUr Diagnostik
D-6200 Wiesbaden, Federal Republic of Germany
^Battelle Institut e.V.
D-600 Frankfurt/M. , Federal Republic of Germany
We examined the prostate with ultrasonics to diagnose tumorous alterations of the
organ. We scanned directly from the abdominal wall through the filled bladder (trans-
vesical). The research is aimed at: 1) improvement and mathematical preparation of
the ultrasonic pictures to assist the physician in his diagnosis; and 2) ascertainment
of significant parameters which allow the classes "normal prostate," "adenoma" and
"carcinoma" to be distinguished in screening examinations. The results show that with
ultrasonics, adenomas and carcinomas are recognizable in 96 percent of the examined
patients and can be differentiated from normal prostate tissue. The palpation finding
was used as a reference in most cases.
Key words: Pattern recognition; prostate tumors; ultrasound.
1. Introduction
Carcinoma of the prostate ranges third in cancer
mortality in the Federal Republic of Germany. Be-
sides the subjective rectal palpation findings which
give the physician signs of organic alterations, it
is useful to look for other examination methods.
The aim of this investigation is to analyze the
value of the ultrasonic scan for differential diag-
nosis and treatment control of tumor diseases and
to check in establishing preclinic findings the
possibility of a screening procedure allowing a
guick separation between suspect and non-suspect
cases in large groups of patients.
The apparatus available for the methodical part
of the investigations are two ultrasonic scanners:
Combison II Kretz (Compound-Scanner) Vidoson, Sie-
mens (Realtime-Scanner) as well as a computer (PDP
11/34). The apparatus is shown schematically in
figure 1. This study includes both examinations in
vivo and some examinations in vitro. Transvesical
representation of the prostate was chosen for in
vivo examination. The ultrasonic pictures are
taken from above the organ by scanning in sections
of 3 mm; the angle at which the sound waves enter
is either « = 15° or a = 20° from the vertical.
Figure 2 shows the position of the organ as well
as the direction of entry of the sound used to pro-
duce the pictures discussed below. The freguencies
are 2.25 MHz and 4 MHz. Linear scanning is used
to cover the area of the organ with a slightly al-
tered angle of entry for the forward and reverse
runs. The B pictures gained in this way are re-
corded on a video tape which serves as mass storage
For the computer analysis the video signal can be
digitized by a specially developed interface and
read into a 512 x 512 matrix; a region of interest
can be marked out within the picture. In addition,
it is possible to feed the non-demodulated HF sig-
nal or the A signal into the PDP 11 computer. Two
control techniques were employed in the analysis of
the ultrasonic pictures: comparison with the urolo
gists' palpation findings and in vitro examinations
of dissection material taken during operations.
The ultrasonic pictures reproduce the geometri-
cal arrangement and form of the density gradients
in the tissue which are situated perpendicular to
the direction of the sound. Since not enough is
known about the reflecting properties of the vari-
ous tissues, the picture features for the different
picture classes cannot be defined exactly a priori .
The study was therefore carried out in the follow-
ing stages:
1. Definition of a limited learning sample
which can be diagnosed as certainly as possible
with the palpation findings and histological
control s .
2. Determination of as many apparently signifi-
cant features as possible.
3. Generation of an adaptive classifier which
begins with a fixed initial weighting of the
features. This weighting is changed in the course
of the classification of the learning sample so as
to eliminate errors. In this way the physician
297
Fig. 3. Ultrasonic picture of a normal prostate.
Fig. 4. Ultrasonic picture of an adenomatous
prostate.
Ultrasonic picture of a carcinomatous
prostate.
supervises the determination of features and class
1 imits.
4. Ascertainment of the validity of the indi-
vidual features to determine a precise set of
distinguishing features which does not involve too
difficult or costly application. The criterion for
validity is provided by the alteration in detection
rate for the relevant feature.
Figures 3, 4 and 5 each show one representative
of the classes normal, adenomatous and carcinomatous
respectively. While the normal organ's interior is
characteristically largely echo free, the adenoma
appears highly structured. The carcinoma in the
right half causes a "shadow" which penetrates the
capsule areas. Other features of the pictures will
be discussed below.
2. Preprocessing of the Pictures
Preprocessing the pictures is intended to allow
the doctor to make the most certain possible diag-
nosis. For this it is necessary to eliminate dis-
turbances and distortions in the picture and to
create the possibility of accentuating features re-
levant to suspected cases, if this is required. To
this end the operations described below were im-
298
plemented on the digital computer and then first
tested by subjecting them to the same classifica-
tion procedure as the original pictures. The fol-
lowing operations were carried out on the pictures
(for different numbers of patients in each case).
a. Definition of a "region of interest." This
operation consists in limiting the organ area on
the screen and reducing the amount of data taken
over in the computer from the picture to 128 x 128
picture points (8 bits per point) which thus refer
to the region of interest. This limitation of the
picture is carried out by the physician treating
the case.
b. Standardizing the intensity of the picture.
The space dependent intensity of the picture x(r,s)
in the B scan picture varies a great deal from
patient to patient (age, corpulence, etc.) and also
depends on the amplification of the experimental
device. Therefore, parameters dependent on in-
tensity should only be used in the diagnosis when
x(r,s) is divided by the mean value x(r ,s ) deter-
mined in a defined organ area. The resulting pic-
ture is the basis for several computer operations.
c. Inverse filtering. To eliminate the loss
in lateral focus due to the form H, (r,s) of the
sound beam as far as possible, the values ascer-
tained under water for this space dependent cou-
pling were used to determine the filter charac-
teristic value H.(r,s). The conditions for the
realization of (r ,s)
F(H|^(r,s)) • F(H.(r,s)) = 1
were replaced by
E(H|^(r,s))
= 1
1 + F(H.(r,s))
(b)
for numerical reasons. F characterizes the Fourier
transform according to the two space coordinates
r,s. Figure 5a shows the sectional picture of a
0.5 mm thick wire in the tank of water as well as
its filtered representation with improved radial
focus .
d. Symmetrical filtering. The convolution of
the picture X(r,s) with any optional symmetrical,
and thus phase free transmission function H(r,s)
allows a far-reaching alteration of the pictures,
when one varies the parameters nix, ni2> ^i' ^2 in
Y(r,s) = X(r,s) * H(r,s]
with
H(r,s) = Um^e
Two phase free band pass filters with different
mean frequencies Umj and Um2 were realized for
miBi = m2B2 and B2 > B^. Figures 6b (F) and 5c (G)
show the result of the operation for Um2 > Umj. The
classes can be better distinguished in the filtered
version of the picture (F, Umj) than in the original
picture (cf. Section 4). The second filter opera-
tion (G) is interpreted so that the picture ampli-
tude is approximately proportional to the gradient
of the intensity in the original picture. This
variable has to be taken into consideration at the
boundary of carcinomas.
(c)
Fig. 6. Sectional picture of a wire before and
after inverse filtering (a); band pass
filtered sectional picture of a carcin-
omatous organ with Umi (F) (b); and Uma
(G) (c).
299
e. Phase dependent filters. It is easier to
judge the pictures, if the function H(r,s) is not
symmetric, i.e.:
H(r,s) = Umie-(^'"^')/Bf . ^„^,-i(r-r,)^ns-s,)^)/4
with Bf<< Bj. In this case the Fourier transform
F (H(r,s)) is complex and a phase shift exists. In
this way a picture is produced which is interpreted
as pseudo three-dimensional when viewed. If H(r,s)
characterizes a differentiating space filter com-
bined with a nonlinear characteristic curve, then
the same effect occurs together with an accentuation
of the intensity modulation. Figure 7 shows two
examples for different H(r,s) (paper in preparation).
All filter operations, insofar as they are linear,
can be combined to one ooeration.
Fig. 7. Filtering with asymmetrical functions
H(r,s).
f. Equidensities . It is easier to analyze the
weak modulations in the area of the organ more
exactly when the low pass filtered original pic-
ture is provided with lines of equal intensity.
Figure 8 shows an example with an increased number
of picture points.
Fig. 8. Carcinomatous prostate represented in
equidensities in a low pass filtered
sectional picture.
It depends on the problem in question which of
the procedures described is best suited for the
physician's diagnosis. Only the F and G pictures
have been tested with the classification procedure
(cf . Section 4) .
3. Feature Extraction
The problem of feature extraction has to be solv
ed in two ways. The reference vectors are deter-
mined on the basis of the physician's palpation
findings and the vectors x- ai"e obtained from the
ultrasonic pictures. The palpation finding takes
into consideration size, consistency, surface state
and simple form parameters [1-5]^. The analysis of
the ultrasonic picture refers to global and local
features which in the first phase are still partly
obtained by visual scrutiny of the ^reen picture.
Should these features prove to be effective in dis-
tinguishing classes, the procedure will be complete
ly automated.
In particular, the following features are quanti
fied:
1. Longitudinal diameter of the organ.
2. Rise in the base of the bladder.
3. Local gaps in the capsule.
4. Coarse parallel fibrous structure in the
interior of the organ.
5. Fine fibrous structure in the interior.
The automatic determination of features 4 and 5
by multiple correlation of the picture with texture
pictures in which the form of the elements is op-
tional (e.g., elliptical) but their location is
statistical, is being implemented at present.
After the pictur'e has been transferred from the
video recorder to the computer, the following param
eters are extracted according to the definition of
the region of interest:
6. Surface of the sectional picture Fg.
7. Autocorrelation (^^^(O'O)
8. The standardized signal power in the region
of interest, i.e., <j)^^(o,o)/x(r,s)2.
^Figures in brackets indicate literature
references at the end of this paper.
300
9. The amplitude density distribution p(x(r,s))
with the relevant first three moments or
central moments respectively
E(x(r,s)), E(x(r,s) - x(r,s))2,
E(x(r,s) - x(r,s))3.
10. Curve of the autocorrelation function along
a line through the suspect zones (j) (r').
^r^r
11. Coherence width and relative extremes of the
autocorrelation function <t)xx('^) along a line
through suspect zones.
12. Power and energy above a threshold or with-
in an amplitude window respectively. This
operation has so far only been implemented
on 60 statistically selected patients.
In order to include local form parameters in the
consideration, the features 6 through 11 were de-
termined in the original picture and in two filter-
ed pictures (F,G) and classified separately. In
this way the efficiency of the filtering in dis-
tinguishing these features was tested. The effect
of picture standardization and parameter 12 have
so far only been studied on 60 patients. The va-
idity of all the parameters mentioned is being
tested with an adaptive classifier.
4. Classification
If is a feature vector, then classification
consists in the division of feature space by class
limits so that
1 Dj(^) i,j = 1,2, ... m, i M
when is correctly classified; D is a discriminat-
ing function [6]. In this project an adaptive
linear classifier was chosen at first with
Di(Zi) = E + ""^w.
k=l ^
The variable W-j designates the kth component of
the reference vector W'l which is generated in p
steps with the aid of a learning sample according
to the rule
-i,6+i ^ -i 6 '^'^ ^i correctly classified
and
-i,6+i " -i,6 ^ "^6 * ^i,6 ''^ i^i incorrectly
classified with
In the case of a Gaussian distortion of the
features, the decision rule minimizes the quadratic
distance between the reference vector and the vector
to be classified. The decisions on a correct or in-
correct classification in the adaptive process de-
scribed above are based on the physician's palpa-
tion findings. After generation of the reference
vector, the pictures were classified in stages.
First, the set of features 1 through 5 was applied
(Part I), then the parameters 6 through 11 for the
original pictures as well as two filtered versions
of the picture (Part II) and then parameter 12 on
intensity standardized pictures for 60 patients.
A classification on the basis of the complete
featureset (I + II + III) will be realized if
feature 12 is determined for all patients. The
class-specific reference-vectors were established
with 25 patients per class in the first stage of
the project. Therefore the following number of
patients can be considered as a test data set.
Part I. After taking palpation findings from
500 patients, the parameters 1 through 5 (Section
3) were extracted and the classes normal (N) and
suspect (adenomatous (A), carcinomous (C)) distin-
guished. The physician's diagnoses were:
a) 97 patients normal
b) 324 patients adenomatous or carcinomatous
c) 79 patients nonspecif ical ly suspect (prosta-
titis, congestion etc.).
If P(N/A + c) characterizes the probability that
a patient classed as normal nevertheless has an
adenoma or carcinoma (false negative) one obtains
the following results for false classification. In
the case of unequivocal diagnosis:
P(N/A + C) = 3.2 % (false negative)
P(A + C/N) = 7.4 % (false positive)
If the nonspecific suspect cases are also taken
into consideration then P(N/A + C) = 7.2 percent.
The increase in error due to group c can be further
lowered with more precise examination. If the total
number of false classifications is considered, the
probability of error is 3.9 percent. In the 3 group
problem the mean error rate increases to 8 percent.
Part II. The classification of the pictures
from 198 patients according to the features 6
through 11 resulted in a mean error of 11.9 per-
cent for the 2 group problem (normal /not normal)
with the original pictures and 10 percent for the
filtered version F and 14 percent for the filtered
version G. First results are now available for
studying the weighting of the individual features
and altered decision strategies but the analysis
has not been completed.
Part III. Using feature 12 presupposes stand-
ardizing the picture intensity. In a set of 60
patients selected statistically the rate of error
for the 2 group problem was 0 percent. The mean
error for the 3 group problem was 7 percent and
P(C/N + A) = 0 percent. This study is at present
being extended to the whole group of patients (500).
An improvement in the classification can be expected.
As the features in the three classification ex-
periments are at least partially independent from
each other, it is to be expected that the rate of
error can be reduced with a combined set of fea-
tures. The systematic study of these combinations,
which are intended to attain a better feature vector
with a few easily calculable components has not yet
been completed.
The further studies are concerned with ascer-
taining a histologically tested learning sample,
determining the sharpest distinguishing parameter
and obtaining a classifier which takes the statis-
tical parameters of the features into consideration.
301
References
[1] Gaca, A., Loch, t. G., Scheiding, U., von
Seelen, W. , and Wessels, G. , Ultraschallunter-
suchungen der Prostata zur Erkennung von Tumor-
erkrankungen. Report BF-R-62. 991-3 (1977)
Bundesministerium fUr Forschung, Bonn, Kennedy-
Allee. (This report contains a comprehensive
list of the relevant literature.)
[2] King, W. , Kieineyer, W. , Mark, R. , Boyce, W. H.,
and McKinney, W. M. , Current status of prosta-
tic echography, J. Amer. Med. Ass. 226 (4),
(1973).
[3] Takahashi , H. and Ouchi , T., Ultrasonic diag-
nosis in the field of urology. First Report,
Japanese Medical Ultrasonics, pp. 7-10,
Tokyo, (1963).
[4] Takahashi, H., and Ouchi, T., Ultrasonic diag-
nosis in the field of urology. Second Report,
Japanese Medical Ultrasonics, pp. 35-37, Tokyo,
T1964).
[5] Watanabe, H., Igari, D. , Tanahashi , Y., Hasada,
K. , and Saitoh, M. , Diagnostic application of
ultrasonography to the prostate. Invest. Urol.
8, 548-559 (1971).
[6] Fu, K. S., Digital Pattern Recognition
(Springer-Verlag, Berlin 1976) .
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed.. National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
RECENT DEVELOPMENTS IN OBTAINING HISTOPATHOLOGICAL INFORMATION
FROM ULTRASOUND TISSUE SIGNATURES
K. Preston, Jr.,^ M. J. Czerwinski,^ M. L. Skolnik,^ and D. E. Leb^
^Department of Electrical Engineering
Carnegie-Mellon University
Pittsburgh, Pennsylvania 15213, U.S.A.
^Ultrasound Laboratory
University of Pittsburgh Medical Center
Pittsburgh, Pennsylvania 15213, U.S.A.
^Department of Medicine
University of Pittsburgh
Pittsburgh, Pennsylvania 15261 U.S.A.
Statistical, signal -analytic techniques may be applied to ultrasonic pulse-echoes
received from organs of the human body. Simultaneous tissue biopsies of these organs
may be sectioned, stained, and measured by the computerized optical microscope using
various image-analytic techniques. This paper reports some preliminary work along
these lines as related to the kidney and liver.
Keywords: Picture processing; signal analysis; ultrasound.
1. Introduction
In the 1950s Wild and Reid noticed that
there appeared to be a correlation between the
properties of the A-scan in mammography and the
histopathology of breast tissue in human subjects.
Later Fry [2] compared the histopathology of ani-
mal tissue (feline and porcine) with the B-scan
echogram. More recently Mountford and Wells [3],
using digitized A-scans from both normal and
cirrhotic subjects, found that specific signal
parameters (amplitude, rise time, etc.) corre-
lated with liver pathology. Ossoinig [4],
Fields et al. [5], and Taylor and Milan [6] made
further observations of B-scan texture, amplitudes
and positions of the peaks in the demodulated A-
scan, and A-scan amplitude probability distribu-
tion functions ("histograms"), respectively.
The advent of high-speed digitizers and inex-
pensive digital computers has now opened this field
to many new investigations. One of these investiga-
tions was commenced by the authors under a grant
from the National Science Foundation (APR75-08154)
in mid-1975. A preliminary report was presented by
Preston[7] at the first meeting of this internation-
al seminar at Gai thersburgh in 1975. The Carnegie-
Mellon University program is being conducted along
the lines of a project in classical pattern recog-
nition in distinction to a project in biological
modeling. The work is being conducted on both
humans and animals in vivo in cooperation with the
^Figures in brackets indicate literature
references at the end of this paper.
Departments of Radiology, Penology, and Pathology
of the University of Pittsburgh Health Center.
Each subject, or more specifically, the target organ
within the subject, is considered to reside in Sig-
nal Space. This space is interrogated in three
ways. A standard B-scan is made using commercial
equipment (Picker Echoview) and, at the same time,
several A-scans are digitized using a special-
purpose, digitizing system (Stahl Research Labs.).
Finally, in addition to the ultrasonogram and the
digitally recorded A-scan, a tissue biopsy is made
of the organ under examination. These interroga-
tions furnish information which constitutes what we
term "Data Space. "
A visual diagnosis is made from the ultrasono-
gram by the staff of the Ultrasound Laboratory and
is combined with the histopathological diagnosis
carried out by the Department of Pathology. This
furnishes information on disease type and severity.
At the same time a numerical analysis is performed
on the digitized A-scan using the statistical meth-
ods described below. These various methods of
analysis of the information in Data Space provide
us with numerical information which is contained in
Measurement Space. These measurements, taken over
a population of patients, are then processed in a
supervised fashion (correlation of measurements
with predetermined disease types) using a standard
discriminant function software package (University
of California BMD07M) to generate points in Classi-
fication Space. Finally, images of the tissue biop-
sy are digitized by the Biomedical Image Processing
Group at Jet Propulsion Laboratory and analyzed
using the Biomedical Image Processing Unit of the
Department of Radiation Health at the University of
303
Pittsburgh. The balance of this paper discusses
A-scan digitizations, A-scan analysis, image digiti-
zation and processing, and reports results to date
in these areas.
2. Materials and Methods
The ultrasound digitizer is the basic unit used
for digitizing A-scans. The digitizer (Stahl Re-
search Labs, model 4036) is a complete digitizing
system rather than a simple analog/digital con-
verter. Its block diagram is shown in figure 1.
The unit requires only two inputs, namely a syn-
chronizing or gating signal and the analog wave-
form to be digitized and stored. The digitizer
runs under control of an internal 15 megahertz clock.
Burst length and skip burst values are entered us-
ing digi switches on the control panel of the model
4036. When a synchronizing pulse is received, the
digitizer proceeds to sample and digitize the input
data at a 15 megahertz rate until the number of
words stored equals the burst length. The unit then
begins counting additional synchronizing pulses un-
til the count has reached the number set for the skip
burst. When this count has been reached, the next
synchronizing pulse causes another burst to be digi-
tized. This continues until the 4096 word memory is
full (each word is one 8-bit byte). At the present
time we are using a burst length of 128 which cor-
responds to 8.5 microseconds of recorded signal
over which the pulse-echoes are received from about
0.65 centimeter of tissue.
The digitizer incorporates a modem and complete
memory teletype (Texas Instruments model 700) having
twin digital cassettes each of which is capable of
storing 250,000 words of digitized A-scan data.
When the user desires to transmit data to a remote
computer, the modem is interfaced via the telephone
system, a dial-up connection is made, and the digi-
tal cassettes are activated and transfer their data
via the communication link. In our present program
the computer used is a Digital Equipment Corpora-
tion PDPIO model KA at Carnegie-Mellon University.
In all experiments to date, the A-scan is digi-
tized directly from the output of the radio frequen-
cy (rf) amplifier. The unit employed for both puls-
ing the transducer and amplifying the A-scan is a
Panametrics model 5055 custom-modified to provide
bandwidth to 20 MHz. In our experiments we have
used the model 5055 at its full gain setting (60 dB)
and with full bandwidth with the interval damping
resistor set to 250 ohms. The organs of interest
have been the liver and kidney for which we obtain
signal levels at the output of the model 5055 rf
amplifier ranging from a few tens to a few hundreds
of millivolts. (The least significant digit of the
digitizer corresponds to approximately 8 millivolts.)
The transducer utilized is a Panametrics custom-
made biopsy transducer whose characteristics have
been measured by the Acoustics Branch, National
Bureau of Standards, and are presented in figure 2.
As can be seen, the resonant frequency is close to
4 MHz with a 25 percent 3 dB bandwidth.
Fig. 2. Transducer 1-way frequency response as
determined by the National Bureau of
Standards .
MODE CONTROL
EXTERNAL TRIGGER
EXTERNAL SYNC
ANALOG DATA
BURST
WORD
AUDIT
15 MHz CLOCK
SKIP BURST CONTROL
0,32, ■•• , 512
BURST LENGTH
1 28,256, ••■4096
BURSTS PER
BLOCK
RECYCLE RATE
MEMORY UPDATE (FIXED RECYCLE)
PITCH
CONTROL
DIGITAL/ANALOG
CONVERTER
UART
SINGLE SIDEBAND
MODULATOR
TI700
TELETYPE
SLOW AGC
MODEM
4096 WORD
HIGH-SPEED
MEMORY
TWIN
DIGITAL
CASSETTES
EAPHONES
DIGITAL COMMUNICATIONS CHANNEL
Fig. 1. Block diagram showing the three modes of operation of the Stahl Research
Laboratory model 4036 digitizer.
304
SAMPl ( IHC( 5 J
Fig. 3. Five signal regimes for the normal liver:
(a) normalized rf waveform; (b) rectified;
(c) video; (d) frequency spectrum; (e)
deconvolved spectrum. Here time samples
are taken each 100 ns with the frequency
domain covering 0 to 5 MHz in 0.08 MHz
intervals.
3. A-Scan Analysis
Five signal regimes (see figs. 3 and 4) are used
in the analysis of the waveform: (1) Direct rf
waveform, (2) Rectified rf waveform, (3) Video wave-
form (the rf envelope), (4) The rf Wiener spectrum,
(5) The selectively deconvolved rf Wiener spectrum.
For regime (5) the frequency spectrum of the trans-
mitted signal is used to normalize the spectrum of
the received A-scan, except at those frequencies
where the uncertainty in determining the transmitted
spectrum is large. Computationally, the selectively
deconvolved spectrum D(a)) is given by
S.(co) / 1 \ / \
D(co) = + S . (t.)
S fo)) \l + c2/ ' \1 + c2/
Fig. 4. The five signal regimes in the same order
as in figure 7 for the normal kidney.
Figure 5 shows a computer generated graph of four
sequential pulse echoes from a phantom consisting
of a multiplicity of polyethylene disks immersed in
water. This figure shows the graphical plot (above)
and (below) a graytone M-scan rendition. The slight
shift in pulse position which is evident in the M-
scan output is an artifact due to synchronization
problems. The pulse consists of about four cycles
at the resonant frequency of the transducer and is,
therefore, about one microsecond in duration. Fig-
ure 6 shows the Wiener spectra of the same four
pulses presented in figure 5.
Fig. 5. Digitally produced A-scan plots and M-mode
graphs of pulse-echoes reflected from a
plane surface.
where S^-(a)) is the spectrum of the direct rf signal,
Sp(u) is the spectrum of the transmitted signal from
which the A-scan echoes are produced, and quantity
c is the coefficient of variation of the transmitted
pulsed (taken over an ensemble of transmitted
(pulses). This quantity is in itself a function
of the frequency o).
Fig. 6. The frequency spectra corresponding to the
pulse-echoes shown in figure 9.
305
Using a burst length of 128 words, the full
memory holds 32 bursts. We are currently employ-
ing two settings of skip burst, namely, 0 and 256.
At a skip burst setting of 0 the entire recording
takes place during 32 pulses which, with the Pick-
er Echoview running at a PRF of 1 kHz, takes 32
milliseconds. The skip burst setting of 256
lengthens the recording interval to 8 seconds.
A comparison of recordings taken at the two skip
burst settings provides a measure of signal sta-
tionarity (see figs. 7 through 14).
\y\A^/^ — --w\/^^^ —
\/\/\/\ — \I^J\/^/^^^ — ^ — — — -^.'v^
Fig. 7. Thirty-two pulse-echoes obtained from a
normal region of the human kidney in vivo.
Echo separation is a 0.001 second.
Observation of figures 12 and 14, which were
taken at the skip burst 256 setting, indicate that
long-term stability of the frequency spectrum does
not exist. This observation raises some doubt as
to the validity of spectral measurements which have
been previously reported in the literature for use
in tissue identification. Most researchers using
spectral signatures have neither considered nor
even discussed short-term versus long-term stabili-
ty problems. In using the A-scan as a tissue sig-
nature, these researchers must remember that all
observations in vivo are of a dynamic system. Even
with held inspiration, the organ in the patient is
moving throughout the cardiac cycle and, in typical
clinical ultrasound laboratories using scanners with
hand-held transducers, further movement is intro-
duced through the inability of the ultrasonographer
to hold the transducer steady.
~'j;^\^\/'^-~-'^~^\f\/^ sA^ ^^^.A^— -^-^N^^vVXy — ^ ^/V^
^-^Aa /V^--^\/Va^- — — — /-vysy^w'-'- ^^\/^--^^>^^^v\A-
Fig. 8, Thirty-two pulse-echoes obtained from a
carcinomic region of the same kidney en-
sonified in preparing the data shown in
figure 11. Pulse-to-pulse separation is
0.001 second.
Fig. 9. The thirty-two frequency spectra correspond-
ing to the A-scans shown in figure 11.
306
--W^/\/\a.^^yV5{/\/\A/V/%^ --v-^V— -^^^
— \/\y\/\/^^v — v/\yV--N/\^ — ^yAyvx/^.-^^^ — - — ^- — ■
^''^-^-^YA-yv — —^-7 — — ^XXAA/^-^^"^ ^~ — ^ — ~\^^\yx^v^\/v
— — - — -^yY\7V'-V\/\AA^^-~-'^^wV\^~ — .^v — ^
Fig. 10. The thirty-two frequency spectra corre- Fig. 12. Thirty-two pulse-echoes obtained from
spending to the A-scans shown in figure 12. a carcinomic region of the same kidney,
again with a pulse-to-pulse separation
of 0.25 second.
Fig. n
Thirty-two pulse-echoes obtained on a
long-term basis from a normal region of
the same kidney ensonified in preparing
the data shown in figure 11. Pulse-to-
pulse separation is 0.25 second.
2.W1HI 3,9nHi
Fig. 13. The thirty-two frequency spectra corre-
sponding to the A-scans shown in figure 16.
307
Fig. 14. The thirty-two frequency spectra corre-
sponding to the A-scans shown in figure 16.
4. Data Reduction
In the early 1970s Mountford and Wells [3,8,9]
described their work in the quantitative analysis
of A-scans taken from normal and cirrhotic livers.
Initially their data was obtained from oscilloscope
photographs of the rf A-scan obtained from the sub-
coastal examination of supine patients through the
anterior abdominal wall (the same physiological ap-
proach taken in our own work). Thirty A-scan oscil-
lograms were recorded per patient over a population
of 30 normal and 13 cirrhotics using an ultrasonic
frequency of 1.5 MHz (with an unrecorded bandwidth).
One hundred data points were recorded for each A-
scan over a time-span of 65 microseconds which cor-
responds to approximately 4.8 centimeters of tissue
(digitizations started at approximately 5 centi-
meters beyond the anterior abdominal wall). Mea-
sures were made which related to signal amplitude,
the rate of change of signal amplitude, frequencies,
rise-times, fall-times, peak-to-peak amplitudes,
etc. The most significant of these measures in
separating the normal from the cirrhotic liver
(using Student's t-test) are tabulated below.
Measure
t-test
Rate of mean amplitude decay
1
28
Mean echo amplitude at intercept
12
38
Mean echo amplitude at mid-point
13
31
Mean trough-to-peak rise-time
4
20
Mean peak-to-trough fall-time
2
39
Mean trough-to-peak amplitude
2
07
Total number of troughs
3
55
Total number of peaks
3
59
As can be seen the most successful measure is the
mean echo amplitude at the mid-point of the time-
window. The associated t-test value indicates an
extremely high success rate in separating normals
from cirrhotics over this particular patient popula-
tion. Other significant measures are the mean rise-
time and the total number of peaks and troughs. It
is somewhat surprising, since these measures direct-
ly relate to the frequency of the signal, that the
authors state that "(the) restricted bandwidth of
the system used in the present study appears to
prejudice the possible use of frequency spectrum
analysis as a diagnostic index."
Subsequent to the above-reported experiment the
authors modified their experimental approach, dis-
carded photographic recording of the A-scan, and de-
veloped equipment for real-time digitization. In
their second experiment 100 time-samples were extract-
ed per patient the sampling interval being 0.1 micro-
second (corresponding to a tissue span of 2 centi-
meters). The initiation point of their digitizations
was 6.5 centimeters beyond the anterior abdominal
wall. Ten normals and three cirrhotics were examin-
ed with 180 A-scans recorded in interrupted quiet
respiration and an additional 180 A-scans recorded
at held maximal inspiration (except for cirrhotics in
which recordings were made only at held maximal in-
spiration). In the second experiment the only mea-
sures reported were the rate of mean echo amplitude
decay and the mean echo amplitude and intercept.
The frequency utilized was 1.67 MHz and it can be
inferred from the authors' discussion that the pulse
duration was approximately 3 microseconds. The sam-
pling aperture was 12 ns with digitization at 6-bits.
Because of the small sample size the t-test was not
calculated, but it was found, as before, that the
mean echo amplitude at mid-point (for held maximal
inspiration) could successfully be used to separate
normals from cirrhotics. Interestingly, the data
spread in this experiment appears to be far greater
than in the initial experiment indicating that the
earlier low error rate values may be subject to
question. After these first two experiments were
conducted, no further investigations apparently have
been carried out.
Our own analysis has gone beyond that of Mount-
ford and Wells in that there is equal emphasis on
both frequency-domain analysis and time-series
analysis. The analysis has been carried out accord-
ing to traditional statistical lines (see, for ex-
ample, Wilks [10]). This approach also appears to
have been taken by Decker et al. [11] at the Uni-
versity of Bonn, in the analysis of A-scans taken
from the eye. In our own work we consider each of
the five regimes of our data as a tissue signature
and, for each A-scan recording for each subject,
make the same statistical measures. The signature
is represented as follows:
Signature = f (x^ )
i = 1,. . . ,128
The function f may represent any one of the five
signal regimes. In other words, it may be either
the rf recording, the rectified rf, the envelope of
the rf, or the Wiener spectrum or deconvolved Wiener
spectrum of the rf.
Measures With Respect to the Argument
Eight measures are made with respect to the argu-
ment X. Before making these measures the function
is normalized in the following manner:
308
= 1
This normalization is applied to both time-series
and frequency-domain data and can be thought of as
a normalization to a standard acoustic pressure for
the time-series data and to a standard acoustic
povver for the frequency-domain data. After normali-
zation sunmations are carried out over four "bins"
of the data points as follov/s:
N
th
■i 3M
N = 1
n = 2
N = 3
N = 4
= 0-15
= 16-31
= 32-47
= 48-63
This produces the first 4 measures for each regime.
Next the moments of f are calculated as follows:
„th
Moment = T. - 0^f{x.) N = 2,3,4
x.f(x.)
The above equations produce 8 measures v/ith re-
spect to the argument x. Next the statistical
properties of the values of f are computed. A
histogram or probability density function of the
values of f is calculated and called p(yj) v/here
p is the likelihood that f has the value yj. The
probability density function is normalized to 0
mean and unit variance as follov/s:
0
1
This normalization once again insures that arbitrary
settings in acoustic amplification and on received
power levels do not effect the results. (Note that,
in contradistinction to the experiments by Mountford
and Wells, we do not operate a system which has
open-loop control on the transmitted and received
signal levels but, rather, operate in an arbitrary
level mode). Once the function p has been computed,
we record its third and fourth moment v/hich, of
course, are proportional to the skew and kurtosis of
this function, respectively.
Then the co-occurrence matrix or diagram is
generated, i.e., p(yj,y|^) which is the likelihood
that two adjacent values of f have precisely the
values yj and y|^. This function is tv/o-dimensional
and may be represented using a scatter plot of the
values yj and y|^ as is shov;n in figure 15. As can
be seen from figure 15, an ellipse may be fitted to
the data from which v/e compute eigenvectors (the
major and minor axes), eigenangle (the angle made to
the yj axis), and the eigenvolume (area of the el-
lipse) .
Finally, the 3-gram is computed. Since there are
only 128 values of f, a three-dimensional scatter-
gram is sparsely filled from a statistical point of
view so that, rather than fitting a three-dimension-
al ellipsoid, an 8-bin histogram is computed with
the values of f inarized with respect to a threshold
equal to the mean value of f so that the bins of the
histogram are given by
(yj.yk^y^) = 000,000,. ..,111
Fig. 15. The di-gram (scatter plot) corresponding
to a single A-scan taken from a normal
liver using 5-bit digitization over 28
points taken at 100 ns intervals.
The range of this binary 3-gram is then entered as
the final measurement. This leads to 8 measures
with respect to the argument x, 2 measures related
to first order probability, 3 measures related to
second order probability, and 1 measure of third
order probability. Since all of these measures are
computed for each of the five data regimes, a total
of 14 measures are made per regime, leading to 70
measurements per A-scan.
5. Biopsy Tissue Image Analysis
As mentioned in Section 1, biopsied tissue is be-
ing prepared not only for visual examination but al-
so computer analysis. Using standard 4 m thick
tissue slices stained with hematoxyl in-eosin and
mounted on microscope slides, the Automatic Light
Microscope Scanner (ALMS) at the Jet Propulsion
Laboratory produces 512 x 512 pixel (picture point)
digitizations of the previously ensonified sample
as shown in figures 16 and 17. Each image field is
approximately 0.5 millimeter square selected so that
a statistically sufficient number of various tissue
cell types are displayed. A software system, called
AUTOPIC, was written incorporating cellular logic
commands for image analysis in the Cartesian tes-
sellation patterned after the hexagonal parallel
pattern transforms of Golay [12]. These transforms
have been found useful by Preston [l3,14] in the
analysis of images of human white blood cells, chest
x-rays, ophthalmol ogical ultrasonograms, etc. Using
AUTOPIC applied to the images shown in figures 16
and 17 normal cell nuclei and i nf 1 anmatory cell
nuclei v;ere located (figs. 18 and 19) using a com-
mand sequence which caused the analysis flow-charted
in figure 20 to be carried out. Since an increase
of inflammatory cells (called "microcytes" in fig.
20) is an indicator of disease, this analysis pro-
vides an index of histopathology which may be cor-
related with the A-scan measures to determine the
relevance of these measures in assaying disease
entities.
309
Fig. 16. Computer generated graytone image
of a 0.5 X 0.5 mm region in an
eosin-hematoxyl in stained tissue
section from a normal rabbit kidney.
Fig. 17. Graytone image (0.5 x 0.5 mm) of
pyel onephri tic rabbit kidney tissue.
Fig. 18. Computer analysis of the normal
rabbit kidney tissue image shown
in figure 16 showing an overlay
of white squares and white circles
to indicate nuclei classified as
microcytes and macrocytes,
respectively.
Fig. 19. Computer analysis of the pyelone-
phritic rabbit kidney tissue image
(fig. 17) showing a microcytes and
macrocytes as in figure 18.
6. Results
Most of the program to date has been allocated to
both the construction and calibration of hardware
and to the development of a software system for both
the computation and graphical display of the data
gathered. In this paper we report on our first
human biopsy patient, a female aged 64 years, whose
pathology was a large carcinoma of the kidney. By
observation of the B-scan the region to be digitized
is readily located. Using the equipment described
in Section 2, the A-scans given in figures 7 through
14 were recorded.
Thirty-two of these A-scans were of normal kidney
tissue and 32 of the carcinoma. Using the short-
term data given in fig ures 7 through 10, all 70 mea-
sures were examined for the 64 recordings and the
two measures most useful in differentiating normal
and pathological tissue were found to be the second
moment with respect to the argument x where x was
selected in the frequency-domain and the contents of
the second bin (or quartile) of the rectified time-
310
GRAY SCALE
IMAGE
COPY
[5] ^
MASK
SELECT 128
WINDOW [l]
_ rBLOCKLj
i] — I — n L
THRESHOLD
FOR DARK
OBJECTS [2]
REMOVE SMALL
DARK OBJECTS
FORM MASK
HFOR LARGE DARK
OBJECTS [5-6]
THRESHOLD FOR
MODERATELY DARK|_
OBJECTS [9]
REPLICATE POSSIBLE
M I CR0CY1ES
[10-45]
MICROCYTES
REPLICATE NON
MACROCYTES
[6 9-88]
DECREMENT AND
CONTINUE REPLI-
COPY
THRESHOLD
CATION OF POSSIBLE
RESULT
AGAIN [48]
MICROCYTES [49-64]
[46]
SUBTRACT
[65-66]
STEREO TYPE
MACROCYTE
RESIDUES [1 31 - I 34]
COUNT
AND
OUTPUT
[136-1 37
COPY 1
SUBTRACT
[I27]t—
[129-130]
REMOVE
NOISE
[68]
REDUCE TO
RESIDUES
[126-126]
ERASE
RESIDUES
[128]
SUBTRACT
REMOVE
REDUCE IN
[89-90]
NOISE [91 ]
SIZE[i08]
THRESHOLD FOR
MODERATELY
DARK OBJECTS [92]
OBTAIN
MACROCYTES
[12 4]
FORM MICROCYTE
MASK [121-123]
COPY
SUBTRACT
COMBINE
REDUCE TO
RESIDUES
[10 3]
f
[1O5-IO6]
t -
[109- 111]
[112-114]
REMOVE
NOISE [95-99]
REDUCE SMALL
-H OBJECTS TO
RESIDUEs[lOO-l02]
ERASE
RESIDUES
[104]
GENERATE
MICROCYTE
STEREO TYPE
[11 5-1 1 7]
COUNT
AND
OUTPUT
[119-120]
Fig. 20. Block flow chart of the image analysis program carried out in the digital
computer analysis of the tissue images shown in figures 16 and 17.
series. Scatter plots of these measures are shown
in figure 21 for the 32 A-scans of both the normal
kidney and the abnormal kidney under short-term con-
ditions. The Mahalanobis Distance was taken as a
measure of separability and indicates a high degree
of confidence in telling these two types of tissues
apart. However, when the A-scan data taken under
long-term conditions (figs. 11 through 14) were
utilized, the Mahalanobis Distance indicated some-
what worse differentiation between the two types
of tissue (fig. 22) typifying the effects of the
dynamic variability of the A-scan.
The other experiment which is reported here in-
volved the introduction of pyelonephritis in the
animal kidney and the comparative analysis of A-
scan data and histologic data derived from images
of biopsies of the ensonified tissue. The animal
model selected was the rabbit.
The disease was introduced unilaterally. At the
acute stage, four weeks after introduction of the
disease, the kidneys (both normal and abnormal)
were operatively exposed and 128 A-scans were re-
corded for each organ via a 6 cm water bath coupler
suggested by Dr. Goans of the Oak Ridge National
Laboratory. This avoided signal corruption by in- -
tervening tissue.
Results of the A-scan analysis (for animal R06)
are shown in figure 23, which is a scattergram of
the data on 256 A-scans taken in groups of 32 in 8
second intervals over several minutes using the
two best computer measures for separating normal
from abnormal. The Mahalanobis Distance is 3.2
and the classification success rate is 98 percent.
Similar results (100% classification) were ob-
tained for the tissue image analysis (fig. 24) using
a single measure, i.e., the count of microcyte den-
sity in the four 0.25 x 0.25 mm regions shown in
figures 17 and 19. Pictorial and A-scan data taken
from six other animals is now being analyzed.
' .0 75.0 150.0 225.0 300.0 375.0 450. 0 525.0
PSM2
104: Tumor vs. Normal (32 ms )
Fig. 21. Scatter plot of 64 points in a 2-coordinate
space representing the best measures for
differentiating ultrasonic pulse echoes
from those received from carcinomic human
kidney tissue in vivo. The coordinates
are the variance of the power spectrum
(PSM2) and the second quartile sum of the
rectified waveform (RED2). The plot is
derived from the short-term data shown in
figures 7-10.
311
Fig.
75.0 150.0 225.0 300.0 375.0 450.0 525.0
PSMZ
104: Tumor us. Normal
22. Scatter plot similar to figure 23 using
data extracted from figures 11-14 showing
A-scan information gathered in the long-
term (8 seconds). Dynamic variance of the
data causes the classification success to
drop to 87 percent for normal human kidney
tissue and to 91 percent for carcinomic
human kidney tissue.
S-
J I I I I
^ Normal
I I Abnormal
55
220
Number of
microcytes
per quadrant
Fig. 24.
Histogram based on the microcyte count per
quadrant (0.25 x 25 mm) in figures 18 and
19 showing 100 percent success in computer-
ized differentiation between normal and
pyelonephri tic rabbit kidney tissue images.
The microcyte count is indicative of in-
flammation and is significantly higher in
pyel onephri ti s .
□ o
log
log(Pr,D4)
n06: Pyalo vs. Normal
Fig. 23. Scatter plot of 256 points using the two
measures which were found to be best for
distinguishing ultrasonic echoes from
normal and pyelonephritic rabbit kidney
in vivo. These measures are the fourth
moment of the power spectrum (PSD4) and
the variance (second moment) of the rec-
tified signal data histogram. Classifi-
cation success for this long-term data
is 100 percent for normal tissue; 93
percent for pyelonephritic.
7. Acknowledgements _
In addition to the workers and colleagues
acknowledged in the text, the authors would like
to acknowledge the work of Dr. Niel Wald, Chairman
of the Department of Radiation Health, University
of Pittsburgh and the staff of his Biomedical Image
Processing Group, Jet Propulsion Laboratories for
assistance in image analysis and recording; Dr.
Donald Eitzen, Ultrasonic Standard Group, National
Engineering Laboratory, National Bureau of Standards,
and his staff for performing the calibration of our
tranducer; Dr. Frank Fry, Ultrasound Research
Laboratories, Indianapolis Center for Advanced
Research, and his students for calibration of the
beam pattern of the transducer and measurements of
its performance using phantoms; Drs. Andrew Dekker,
Denis Borochovitz, Jeffrey D. Hubbard, University
of Pittsburgh Health Center, for visual reading
and reporting on tissue section pathology; Dr.
Terrance Matzuk, consultant, for providing ultra-
sound interface electronics; Mr. Gilbert Arnold,
Mellon Institute, and his staff for providing the
illustrations; Mrs. Tanya Rogers, Department of
E-lectrical Engineering, Carnegie-Mellon University
for typing the manuscript.
312
References
[1] Wild, J. J. and Reid, J. M., Application of
echo ranging techniques to the determination
of structure of biological tissue. Science
115. 226 (1952).
[2] Fry, E., Okuyama , D. , and Fry, F. J., Ultra-
sonic differentiation of normal liver struc-
ture as a function of age and species, Proc.
6th Intern' 1. Cong, on Acoustics, Tokyo (1968).
[3] Mountford, R. A. et al . , Ultrasonic liver
scanning: automatic A-scan analysis. Physics
in Medicine and Biology 17, 559-569 (1973).
[4] Ossoinig, K. C, Quantitative echography -
the basis of tissue differentiation, J. Clin.
Ultrasound 1, 190 (1973).
[5] Fields, S. and Dunn, F., Correlation of echo-
graphic visualizability of tissue with bio-
logical composition and physiological state,
J. Acoust. Soc. Amer. 54, 809 (1973).
[6] Taylor, K. J. W. and Milan, J., Digital A-
Scan Analysis in the Diagnosis of Chronic
Splenic Etilargement, in Ultrasonic Tissue
Characterization, M. Linzer, ed.. National
Bureau of Standards Special Publication
453, pp. 71-80 (U.S. Government Printing
Office, Washington, D.C., 1976).
[7] Preston, K. , Jr., Use of Pattern Recognition
for Signal Processing in Ultrasonic Histo-
pathology, in Ultrasonic Tissue Characteriza-
tion, M. Linzer, ed.. National Bureau of
Standards Special Publication 453, pp. 51-60
(U.S. Government Printing Office, Washington,
D.C., 1976).
[8] Mountford, R. A. and Wells, P. N. T., Ultra-
sonic liver scanning: the quantitative
analysis of the normal A-scan, Physics in
Medicine and Biology 17, 14-24 (1972a).
[9] Mountford, R. A. and Wells, P. N. T., Ultra-
sonic liver scanning: the A-scan in cir-
rhosis. Physics in Medicine and Biology 17,
261-269 (1972b).
[10] Wilks, W., Mathematical Statistics, Wiley
(1963).
[11] Decker, D. , Epple, E., Leiss, W., and Nagel ,
M. , Digital computer analysis of time-
amplitude ultrasonograms from the human eye,
J. Clin. Ultrasound 1 (2), 150 (1973).
[12] Golay, M. J. E. , Hexagonal parallel pattern
transforms, IEEE Trans. Comput. C-18, 733
(1969) .
[13] Preston, K. , Jr. and Onoe, M. , Digital Proc-
essing of Biomedical Images, Plenum Press
(1976). ■
[14] Preston, K., Jr., Application of the Golay
Transform to Image Analysis, in Digital
Image Processing and Analysis, 1' Institute
de Recherche d Informatique et d ' Automatique ,
Paris (1976).
313
CHAPTER 10
TISSUE VIABILITY AND TISSUE PHANTOMS
315
i
!
i,
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed., National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
DAMAGE AND DEATH IN TISSUES AND ASSOCIATED CHANGES IN THEIR MECHANICAL PROPERTIES
L. Weiss
Roswell Park Memorial Institute
Buffalo, New York 14263, U.S.A.
Mechanical effects of tissue damage associated with artefactual change during
examination, and that developing naturally during tumor development, have been de-
monstrated by a quantitated cell detachment test. Although the precise relationship
between this semi-destructive test of tissue properties and ultrasonic, nondestruc-
tive tests is uncertain, it appears feasible to consider the following points In
studies of the interaction of ultrasound with tissue specimens: a) thin specimens
should be examined quickly to avoid anoxic and hypoxic damage; b) note must be made
of the heterogeneities within solid tumors, particularly those produced by necrosis;
c) degenerative changes in a tumor and surrounding non-malignant tissues may act as
image-enhancers .
Key words: Cell detachment; mechanical properties; tissue damage; tissue death.
1. Introduction 2. Cell Detachment from Tissues
The interactions of ultrasound with tissues. Cylinders are punched from tissues using a 13
depends in a sense on their mechanical properties. gauge ("^ 2 mm) trocar and cannula, and these are
then divided into 1 or 2 mm lengths over a scale,
under magnification. Two such cylinders are placed
into each of a number of screw-capped, glass vials,
4.5 cm long and 1.3 cm diameter, containing 2 ml
of cold Hanks' balanced saline solution. These
operations, which take 20 to 30 minutes, are all
done in solutions at 4 °C to minimize metabolic
di fferences .
The vials are clamped onto a reciprocating
shaker, making 275 oscillations per minute, with
an excursion of 4.5 cm, for 40 minutes, at room
temperature (25 °C). In some experiments, tissue
cylinders were shaken for 10 minutes only. After
shaking, 0.25 ml of 10 percent buffered formalde-
hyde is added to each vial and mixed by gentle
inversion. When present, the macroscopic remains
of the cylinders rapidly settle to the bottoms of
the vials, and the released cells and cell clusters
in the supernatant fluid are counted in a Fuchs-
Rosenthal chamber, and their diameters measured
with an image-splitting eyepiece (Vickers-A. E. I . ,
England) at a final magnification of X300. It has
previously been demonstrated that the described
formaldehyde-fixation does not affect these two
numerical determinations.
An estimate of the volume of tissue liberated
by shaking was obtained from the product of the
cube of the mean "particle" radius and the mean
number of such cell units. Volume functions of
this type are used for purposes of comparison.
Small pieces of tissue were fixed in formalin
and subsequently examined in 4 ym stained sections.
Aliquots of liberated cells are centrifuged onto
In this presentation, tv;o facets of tissue damage
will be examined from a mechanical viewpoint.
The first is a consideration of the hypoxia which
results and its possible artefactual effects, when
pieces of tissue are examined. The second is a
consideration of the mechanical heterogeneity in
and around solid cancers.
The mechanical properties of tissues depend
partially on the properties of their individual
cells and partially on the properties of the
material lying between them, in much the same way
as the properties of a brick wall depends on the
properties of the bricks, the mortar and the bonds
between them. Tissue properties may be measured
by nondestructive techniques, such as those in-
volving ultrasound, or by destructive techniques
which result in reduction of the tissue into either
subcellular dimensions, or into its component liv-
ing cells. The release of living cells from tis-
sues have been studied, particularly solid cancers,
as part of a program on the mechanisms of metas-
tasis [1-3]^. In this work, quite sensitive
quantitative techniques have been developed to
measure cell detachment, which is dependent on the
strength of the intercellular material. In tissues,
the intercellular material represents a modifica-
tion of the extramembranous , peripheral regions of
cells which is rich in glycoproteins.
^Figures in brackets indicate literature
references at the end of this paper.
317
slides (Cytospin: Shandon-Ell iott, England), and
morphologically characterized after reaction with
Wright's stain. A full description of the tech-
nique and its use is given elsewhere [4].
The cell detachment test described here is non-
destructive as far as the component cells of a tis-
sue are concerned, and their viscoelastic properties
must play a role in cell detachment. However, as-
suming that the plane of detachment lies within the
cell periphery/intercellular region, then the tests
are destructive for this tissue component in the
sense that they are essentially concerned with ob-
taining a "yield-point". In contrast, the ultra-
sonic techniques discussed here are completely non-
destructive. Thus, although the precise relation-
ship of the two techniques is problematical and the
results are expected to be somewhat different in
general terms it would be expected that changes in
tissues detected by one procedure might well be re-
flected in the other. Proposed work will hopefully
clarify this relationship, which at present is
simply presented as a working hypothesis.
3. Examination Artefacts
In determining parameters of the interactions
of ultrasound with tissues, the tissues them-
selves are held in a tank of fluid in the ultra-
sonic beam, for varying periods of time. From
many biologically oriented studies, it is well-
known that unless the fluid is of physiologic
tonicity and pH, and that unless the tissues them-
selves are well -oxygenated , tissue damage will
occur. The relevant question is, how much arte-
factual alteration will be produced in the ultra-
sound/tissue interaction by this damage?
From a practical viewpoint, it is unnecessary
to investigate the effects of unphysiologic fluids
on the interactions, because many solutions are
available which maintain optimal conditions of
ionic strength and constitution, pH etc. [5].
However, one outstanding problem is the possibility
of inadequate tissue oxygenation, and the con-
sequences of it. In the case of isolated tissue
slices, Warburg [6] showed that their limiting
thickness in cm (H) for O2 consumption, where D
is the diffusion coefficient, A is the rate of
respiration of the tissue (ml O2 uptake/ml
tissue/minute), and [O2] is the oxygen concentra-
tion in the atmosphere surrounding the tissue,
then:
H = (8-D-[02]/A)''= .
Thus, in pure oxygen, the maximum thickness of tis-
sue permitting the passive penetration of O2 to its
center is 0.47 mm. In air ([O2] = 21 percent) the
corresponding maximal thickness of tissue is only
0.21 mm. Similar considerations have to be taken
into account in determining the depth of fluid
covering examined tissues. Studies on the kinetics
of gas diffusion in cell culture systems by Mc-
Limans et al . [7] have demonstrated that as pre-
dicted by Fick's Law, the passive diffusion of
oxygen through unstirred fluid to actively metabo-
lizing tissues is very sensitive to the depth of
fluid, and that in the case of isolated liver cells
covered by only 1.6 mm of physiologic saline, the
total initial equilibrated concentration of oxygen
will be exhausted in only 1280 seconds, normal bio-
logic syntheses will be impaired and, if left under
these conditions, the cells will die.
In order to get a feeling for ti ssue-oxygenation
under the conditions in which ultrasonic tissue
characterization experiments will be made, freshly
isolated cylinders of rat liver 1 cm in diameter
and 0.5 cm in length, were obtained from exsan-
guinated rats. These were then placed in a beaker
containing 100 ml Hanks' balanced saline solution,
a commonly used mammalian "physiologic" solution,
buffered at pH 7.2, and maintained at 37 °C. The
PO2 was measured at various depths in the liver
cylinder determined with a micrometer, with oxygen
ultramicroelectrodes having a tip diameter of
15 pm. The electrode has been described elsewhere
[8]. It was shown that in the five minutes re-
quired to start measurements after removal of the
liver cylinders, the oxygen tensions at depths of
1 to 2.5 mm were near zero, and at a depth of 0.5
mm, they were approximately 60 mmHg. These
measurements compare with mean values for the liver
in the unanesthetized, spontaneously respiring rat
of 20 to 30 mmHq [9]. Thus, there is good reason
for considering the effects of hypoxia in ultra-
sonic experiments, since in most of these, com-
paratively thick pieces of tissues are used.
It has been shown that lysosomal changes in
rat hepatic parenchymal cells are induced by sub-
jecting animals to hypoxia [10]. It has also
been shown that activation of lysosomes with re-
lease of lysosomal enzymes into the tissues can
cause massive tissue degradation [11] within
hours, and changes detectable by cell detachment
tests on cellular monolayers, within 2 minutes
[12].
In order to determine if the treatment of the
1 cm diameter cylinders leading to anoxia or
hypoxia resulted in analogous changes, shaking-
tests of the type described earlier were made on
this material. After incubation in Hanks' solu-
tion at 37 °C for periods up to 2 hours, plugs
were cut from the 1 cm cylinders, and the volumes
of cells released at the different times, from the
outermost 1 mm regions near to the cut surface and
the innermost 2 mm thick regions at the center of
the 5 mm thick cylinder were compared. The re-
sults shown in table 1 indicate that greater
(.02 > p > .01) volumes of parenchymal cells are
released from the inner anoxic region at the
center of liver cylinders than from the outer
parts. During the experiments, detectable amounts
of protein and acid phosphatase (a lysosomal hydro-
lase) are released into the medium, but this facet
of the experiments needs developing.
It should be noted that examination of standard
histologic preparations of the liver revealed no
clear-cut evidence of damage after 2 hours incuba-
tion, and would not have permitted predictions of
the detachment experiments. It is of interest
that Hueter, in experiments cited by Dunn [13] was
unable to detect changes in the ultrasonic absorp-
tion coefficient of liver, measured over the range
of 1 to 6 MHz, until 9 hours after death, when it
progressively decreased. It is difficult to com-
pare these experiments with those reported here,
since Hueter 's were made at 25 °C (c.f. 37°) and
the tissue was maintained at 10 °C between measure-
ments, and it is well-known that reduction in tem-
perature retards post-mortem change. Future work
will hopefully clarify the correlation between
changes detected by the shaking test and ultra-
sound.
If degenerative, autolytic changes due to
lysosomal mobilization prove to be an inconvenience
318
Table 1. Release of liver parenchymal cells.
Incubation Sites Numbers of cells Unit diameters Comparative
time released (xlOOO) (pm) + SE volumes released
(minutes) ± SE (n) (n) Inner : Outer
Experiment 1
30
Inner
57
+
7.
3
(13)
10.5
+
0.
14
(161)
1
1
8
30
Outer
37
+
4.
9
(16)
10.0
+
0.
13
(110)
60
Inner
20
+
3
5
(14)
10.0
+
0.
15
(131)
1
1
4
60
Outer
15
+
1
5
(13)
9.9
+
0
13
(107)
120
Inner
16
+
0
7
(8)
10.5
+
0.
39
(100)
1
1
5
120
Outer
12
+
1
9
(12)
10.0
+
0
27
(105)
Experiment 2
30
Inner
16
+
2
8
(8)
10.0
+
0
11
(102)
1
1
9
30
Outer
10
+
2
7
(9)
9.5
+
0
11
(94)
60
Inner
15
+
4
0
(6)
9.6
+
0
13
(136)
1
1
2
60
Outer
13
+
3
4
(6)
9.4
+
0
08
(154)
120
Inner
13
+
1
9
(9)
9.9
+
0
09
(111)
1
1
.5
120
Outer
10
+
2
4
(8)
9.4
+
0
12
(98)
in ultrasonic measurements, it is worthy of note
that the analogous changes observed in vitro [11,
12] were inhibited and/or retarded by adding small
quantities (c. 10 yg/ml ) of hydrocortisone hexi-
succinate to the medium.
4. Mechanical Heterogeneity in Tumors
The essence of the successful treatments of
cancer is early diagnosis. In the breast for ex-
ample it can be seen (fig. 1) that depending on
the clinical type [14], the probability of metas-
tasis varies considerably with the size of the
cancer at the time of diagnosis. In the most
malignant (Type A) the slowest rate of increased
probability of metastasis, where therapeutic in-
tervention would be most effective, is in lesions
of 1 to 7 mm diameter. In the least malignant
(Type B), there is an increased probability of
metastasis from 22 to 36 percent, as the lesion
increases in diameter from 1 to 5 mm. It is there-
fore pertinent to ask whether changes in or around
a cancer as it develops can so modify the mechani-
cal properties of the cancer itself or the sur-
rounding nonmalignant tissues, that the image of
the tumor is enhanced and/or changes in the sur-
rounding tissues effectively increase the "lesion"
size.
From previous considerations of examination of
artefacts, the question arises of whether various
tissue interactions caused by, or resulting in de-
generative change in a solid cancer could fill this
role of "image enhancer". Cell death, or necrosis
is a salient feature of many solid cancers. One
contributory cause of this is that blood vessels
entering or leaving cancers from their peripheries,
tend to be occluded by tissue pressures associated
with growth, leading to vascular stasis and throm-
bosis. The inward diffusion of nutrients, and the
outward diffusion of metabolic products becomes
inadequate, and tissue death results. There are
other contributory causes of necrosis in and around
tumors; including cell and humorally mediated
cytotoxicity associated with inflammatory change,
but the end result is the same. In a cancer with
5
4-
3 -
1
0.8
0.6
0.2
O.lL
type B ,
type A
Fig. 1.
0 20 40 60 80 100
Probability of distant metastasis
The probability of distant metastases in
breast cancers of types A and B, of dif-
ferent size at the time of diagnosis.
319
well-developed necrosis, the necrotic material
liquefies and the lesion becomes cystic. The dif-
ferent absorptive properties of the liquefied and
solid regions lead to image enhancement--as is
well-recognized on a macroscopic scale, and rep-
resents an example of the heterogeneity of tumors
[15].
Much previous work has not taken into account
the heterogeneity of normal , and particularly
pathologic tissues. This heterogeneity may be due
to normal anatomic structures such as large blood
vessels, fascial planes etc. , or to local patho-
logical situations such as necrosis, cyst forma-
tion, cellular infiltrations, tissue edema etc.
Very few pathologic lesions consist of a "pure"
tissue type, and for this reason average, integrat-
ed interaction data collected from a large volume
of tissue may be inappropriate for tissue signa-
ture studies. In any general consideration of
ultrasonic tissue characterization it is mandatory
to consider not only the macroscopic or average
features of lesions, but also their individual
"microscopic" components. By "microscopic", a
region of a lesion is implied which has the dimen-
sions of the narrowest ultrasonic beam used for
tissue characterization experiments. With this in
mind, the mechanical properties of different
regions of the same tumor, and the nonmalignant
tissues around it have been determined by the cell
detachment technique, on Walker 256 "carcinosar-
comas" transplanted in livers of rats.
A. Effect of Site Within Tumor
Cystic, subcutaneous Walker 256 tumors were
selected, with necrotic centers of approximately
3 cm mean diameter, and with wall thicknesses of
2 to 5 mm of "healthy" cancerous tissue. The re-
lease of cells from the inner and outer 1 mm
lengths of cylinders taken radially through the
cyst walls are shown in table 2. In the two rep-
resentative examples shown, 3 and 6 times greater
volumes of tumor were released from the inner,
juxta-necrotic regions of the cyst walls than from
their outer parts.
B. Effects of Necrotic Extract
on Tumor Cel 1 Release
The subcutaneous tumors from which material
was obtained fell into two groups. The first were
roughly spherical, with diameters of approximately
1 cm, and contained minimal, friable necrotic
cores. The second group, were roughly egg-shaped
with diameters of approximately 4 cm and 6 cm in
Table 2.
Tumor cel 1 s
released from inner and outer
parts of walls of subcutaneous
cystic
Walker 256
tumors .
"Cel 1 "-
1 c I era J c
1 itrdil b I Zc
Vol ume
(X 1000)
± SE (n)
ym ± SE (n)
ra 1 1 0
Inner
Outer
Inner Outer
Inner/Outer
1 mm
1 mm
1 mm 1 mm
202 + 65
77 + 12
7"7j.O CT 1 A ^ ") 1
11 ±J.D /4±J.I
3:1
(9)
(9)
(99) (99)
90 + 26
36 ± 8.2
110 + 3.3 83 ± 3.5
6:1
(9)
(10)
(100) (100)
the short and long axes respectively, and were
largely necrotic, with healthy cancerous "rims"
varying from 2 to 4 mm in depth.
Cylinders obtained from the peripheral regions
of the tumors were first incubated with 1/10 dilu-
tions of necrotic extracts in HBSS for 20 minutes
at 37 °C and then shaken for 10 minutes at room
temperature (24 °C). The representative results
of 6 separate experiments with 3 separate batches
of necrotic extract are summarized in table 3.
In the case of the 1 cm diameter tumors, 8 to 11
times greater volume of tumor was released after
exposure to necrotic material than in the appro-
priate controls. In the case of the largely
necrotic tumors, it is seen that pretreatment
with necrotic extract produced no significant
increase in the volume of tumor released by shak-
ing, compared with controls.
C. Effects of Tumor on Surrounding
"Normal" Tissues
Following the direct injection of Walker
ascites cells (10^ cells in 0.2 ml HBSS) into the
liver, tumors of approximately 1 cm diameter grew
within 10 days. The tumors and the surrounding
liver were bisected, and cylinders of liver were
removed at its junction with the tumor, and at
0.5 and 1.0 cm from this interface. The results
of 10 separate experiments, together with re-
lease data from 2 normal livers are given in table
4. It is shown that a greater volume of paren-
chymal cells are released from tumor-bearing than
normal livers. In addition the closer the normal
liver samples are to the tumor interface, the more
readily are cells detached from them.
Table 3. Tumor cylinders incubated with necrotic extract
for 20 minutes at 37 °C; then shaken for 10 minutes.
Exper imenta'
(E)
Control
(0
Tumors
Numbers
CI ump
Numbers
Clump
Volume
rel eased
diameter
rel eased
diameter
ratio
(X 1000)
(ym)
(X 1000)
(ym)
E:C
± SE (n)
+ SE (n)
± SE (n)
± SE (n)
c. 1 cm
171 ± 56
57 ± 2.0
245 ± 16
23 ± 0.3
10:1
(17)
(100)
(17)
(100)
> 4 cm
376 + 56
23 ± 1.0
477 ± 61
21 ± 0.6
1.0:1
(10)
(199)
(10)
(197)
320
Table 4. Release of "normal" liver parenchymal
cells surrounding tumors as function
of distance from tumor interface.
Distance
from
tumor edge
Numbers
released
(X 1000)
± SE (n)
CI ump
size
(ym)
± SE (n)
Vol ume
ratio
(cf. normal)
0 cm
324 ± 16
(65)
42 ± 0.6
(510)
4.1:1
0.5 cm
256 ± 20
(36)
43 ± 0.6
(306)
3.7:1
1.0 cm
224 ± 17
(19)
38 ± 0.5
(204)
2.3:1
Normal
liver
350 ± 16
(84)
25 ± 0.2
(400)
1.0
D.
Effects of Necrotic Extract on
Liver Cell Release
Standard tissue cylinders were obtained from
the livers of normal rats. These were incubated
for 20 minutes at 37 °C, in either 10 percent
necrotic extract in HBSS, or in BSA control solu-
tions. The cylinders were shaken for either 10 or
40 minutes. Representative results of 4 separate
experiments given in table 5 show that after 10
minutes shaking, 3 times greater volume of cells
were released from liver cylinders pretreated with
necrotic extract than from the control series;
after 40 minutes shaking, the ratios of extract-
treated to controls increased to 50:1.
The results of experiments (A) through (D) show
clear-cut differences in the mechanical proper-
ties of different parts of the same tumor, and in
the tissues surrounding them. The changes appear
to be related to the presence and diffusion of
necrotic material. If the increased detachment
patterns is due to liberated lysosomal enzymes in
which the necrotic material is rich, then it must
be remembered that some of these may be contribut-
ed by macrophages and polymorphs [16] which are
frequently found in association with tumors and
necrotic tissues.
5. Conclusions
If these results, based on cell detachment, are
indicative of changes detectable by ultrasound,
and studies presently underway will hopefully
clarify this, then three positive suggestions come
from this work:
A. Attention must be given to the microen-
vironmental conditions under which specimens are
maintained while being examined, and the time
spent on examination must be minimized.
B. It is not enough to determine the charac-
teristics of whole tumors; individual regions
must be characterized and identified at a "micro-
scopic level".
C. Changes consequent to degeneration, occur-
ring in a cancer, and in the nonmalignant tissues
surrounding it, may enhance its detection by ultra-
sound by creating new differentials within these
tissues.
Acknowledgments
My thanks are due to Dr. H. Bicher and Mr. L.
D'Agostino, Department of Radiotherapy for making
measurements of oxygen-tension, and to Ms. J.
Holmes, D. Lombardo and Mr. D. Graham for their
technical assistance.
This work was partially supported by Grants
#PDT-14 from the American Cancer Society Inc.
and CA-17609 from the National Institutes of
Health.
Table 5.
Release of parenchymal cells
from normal
liver ± nectotic
extract.
Experimental (E)
Controls (C)
Time
Numbers
Clump
Numbers
Clump
Volume
shaken
released
diameter
rel eased
diameter
""atio
(X 1000)
(pm)
(X 1000)
(ym)
E:C
± SE (n)
± SE (n)
± SE (n)
± SE (n)
10 m
88 ± 4.4
31 + 1.2
60 ± 3.5
19 ± 0.9
3:1
(10)
(137)
(9)
(127)
40 m
250 ± 4.9
51.3 ± 2.9
40 ± 4.8
25.6 ± 1.6
50:1
(10)
(100)
(10)
(100)
References
[1] Weiss, L. , The Cell Periphery, Metastasis
and other Contact Phenomena (N. Holland
Press, Amsterdam, 1967).
[2] Weiss, L. , ed.. Fundamental Aspects of
Metastasis (North-Holland/American Elsevier,
19761::
[3] Weiss, L., Cell detachment and metastasis,
Gann (1977a) (in press).
[4] Weiss, L., Tumor necrosis and cell detachment,
Internat. J. Cancer (1977b) (in press).
[5] Altman, P. L. and Katz, D. D., eds.. Cell
Biology, p. 61 et seq. , F.A.S.E.B., Bethesda,
Maryland (1976).
[6] Warburg, 0., Ober den Stattwechsel der
Tumoren (Springer, Berlin, 1956).
[7] McLimans, W. P., Crouse, E. J., Tunnah,
K. B., and Moore, G. E., Kinetics of gas
321
diffusion in mammalian cell culture systems.
1. Experimental, Biotech. Bioeng. 10, 725-
740 (1968).
[8] Cater, D. B., Silver, I. A., and Wilson,
G. M. , Apparatus and technique for the
quantitative measurement of oxygen tension
in living tissues, Proc. Roy. Soc. Lond.
(Series B) 151, 256-276 (1959/1960):
[9] Gornandt, L. and Kessler, M. , PO2 Histo-
grams in Regenerating Liver Tissue, in
Oxygen in Tissues, M. Kessler, D. F. Birnley,
L. C. Clark, D. W. Lubbers, I. A. Silver,
and J. Strauss, eds., p. 288 (Urban &
Schwarzenberg, Miinchen-Berl in-Wien , 1973).
[10] Ericsson, J. L. E., Mechanism of Cellular
Autophagy, in Lysosomes in Biology and
Pathology, J. T. Dingle and H. B. Fell,
eds.. Vol. 2, pp. 345-394 (North-Holland,
Amsterdam, 1969).
[11] Fell, H. B. and Weiss, L., The effect of
antiserum, alone and with hydrocortisone,
on foetal mouse bones in culture, J. Exptl.
Med, 121, 551-560 (1965).
[12] Weiss, L. , Studies on cellular adhesion in
tissue culture. VIII. Some effects of
antisera on cell detachment. Exptl. Cell
Res. 37, 540-551 (1965).
[13] Dunn, F. , Ultrasonic Absorption by Bio-
logical Materials, in Ultrasonic Energy,
E. Kelly, ed. , pp. 51-65 (Univ. of Illinois
Press, Urbana, 1965).
[14] Slack, N. H., Blumenson, L. E., and Bross,
I. D. J., Therapeutic implications from a
mathematical model characterizing the course
of breast cancer. Cancer 24, 960-971 (1969).
[15] Blumenson, L. E. (1972) Cited by L. Weiss,
Some Patho-biological Considerations of
Breast Cancer by Ultrasonic Holography, in
Ultrasonic imaging and holography, G. W.
Stroke, W. E. Kock, Y. Kikuchi, and J.
Tsujinchi, pp. 567-585 (Plenum Press, New
York, 1974).
[16] Davies, P. and Allison, A. C, The Secre-
tion of Lysosomal Enzymes, in Lysosomes in
Biology and Pathology, J. T. Dinqle and
R. T. Dean, eds.. Vol. 5, pp. 61-98 (North-
Holland, Amsterdam, 1976).
322
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer, ed. , National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979)
A HUMAN ABDOMINAL TISSUE PHANTOM
P. D. Edmonds, Z. Reyes, D. B. Parkinson
Stanford Research Institute International
Menlo Park, California 94025, U.S.A.
R. A. Filly ;
University of California Medical Center
San Francisco, California 94122, U.S.A.
H. Busey
Picker Corporation
Northford, Connecticut 06472, U.S.A.
The objective of this work was to determine the feasibility of constructing a
phantom that would simulate human abdominal tissues when interrogated by advanced
B-scan diagnostic ultrasound equipment operating at 2.25 and 3.5 MHz. Satisfactory
results were obtained with gelatin based components having dispersed scatterers
and embedded sponge, plastic tubes and rubber bulbs filled with saline solution.
Stability of the gelatin was achieved by addition of a stabilizing agent, a pre-
servative and impervious coatings to prevent water evaporation.
Key words: Gelatin; human abdominal tissues; phantom.
In the course of producing and installing ultra-
sonic diagnostic equipment, a need arises for a
phantom object with stable acoustic properties,
that will generate displays suitable for optimizing
the adjustment of various pre-set equipment levels,
particularly in the grey-scale circuitry. The need
extends beyond the concerns of a single company be-
cause there is a lack of basic standardization in
the industry. The general availability of a suit-
able phantom would improve uniformity of displays
obtained with different ultrasonic imaging systems.
This paper describes the evolution of a phantom
object intended to simulate many of the ultrasonic
properties of human abdominal tissue when imaged
by a Picker Echoview System 801 Laminograph.
Figure 1 is an abdominal B-scan showing the variety
of textures to be simulated.
The design specification called for:
• attenuation coefficient at 2.25 MHz =
1.7 ± 0.1 dB/cm;
• velocity of propagation at 22 °C (72 °F) =
1540 ± 50 m/s;
• simulation of fine and coarse texture of
normal and pathological liver, as well as the
specular reflection of capsular interfaces
and blood vessels, cysts with or without
sediment, and abdominal aorta, utilizing the
full range of grey levels;
• density = 1.05 +0.1 g/cm^;
• mechanical stability up to 65 °C (150 ^F);
• flexible exterior surface permitting limited
sector scanning;
• abrasion resistant surface in the presence
of mineral oil and aqueous gels.
Fig. 1. Human abdominal B-scan showing variety
of textures to be simulated.
Less importance was attached to reproducing spe-
cific organ geometries or spatial relationships.
Initial work focused on the selection of a
base material. Classes of materials considered
were room temperature vulcanizing (RTV) silicone
rubbers, "pc" rubbers, gelatin gels, 3M® percus-
sion pads and "Hydron®" polymer. In order to
simulate "texture", it would be necessary to in-
corporate scattering centers in the base material
Hence, some part of the overall attenuation co-
efficient, A, of an acceptable simulating mate-
rial would be attributable to scattering and the
residual absorption coefficient, a, of the base
323
material without scattering had to be correspond-
ingly less than the desired overall attenuation
coefficient: a = A - a^, where is the at-
tenuation per centimeter due to scattering.
Measured absorption coefficients and propaga-
tion velocities of the candidate base materials
are given in table 1. The absorption data were
obtained by the substitution method and the
velocity data by the pulse travel time method
[1]^. The techniques of sample preparation as
Table 1. Measured material properties
at 2.25 MHz, 22 °C.
Material
Absorption
coefficient
(dB cm-i)
Propagation
velocity
Notes
General Electric
RTV 602
RTV 515
Goodrich
#35080 "pc"
rubber
3M® percussion
pad, style 3290
"Hydron® " , dry
"Hydron® " , wet
10% gelatin/water
+ 1% chrome alum
Specification
1 .34 ± 0.2
2.55 ± 0.3
2.84 ± 0.11
1.7 ± 0.2
16 ± 2
13 ± 2
<0.5
<0.1
<1.7 ± 0.1
980 ± 10
980 ± 10
1546 ± 3
1490 ± 10
2050 ± 12
1950 ± 12
1539 ± 10
1540 ± 50
0.25% SRC05 catalyst; cured 48 hours at room
■temperature.
10% 615B catalyst; cured 48 hours at room
temperature.
Sample formed from uncured rubber stock by milling
,and molding for 45 minutes at 6.6 kg/m^ and 150 °C.
Measured as received in sheet form.
^The sample from Hydron, Inc. (2-hydroxy ethyl
methacrylate polymer in water) was dried and then
molded at 175 °C under pressure. This produced a
hard, brittle, opaque cylinder which was immersed
in distilled water to swell and soften. The re-
sultant swelling was irregular. Internal bubbles
and interfaces became visible. The attenuation
was greater than 10 dB/cm at 2.25 MHz, 22 °C.
This sample was then recovered by dissolving in
ethanol ; the solution was placed in a dessicator
in the presence of ethanol and dried over a period
of 2 months to yield a translucent 1/8 inch thick
disc. The "dry" sample was this disc equilibrated
j,with air.
A portion of the disc was cut and immersed in dis-
tilled water for 24 hours. It became opaque white
and evidently different from contact lens material.
This was the "wet" sample.
^Photographic gelatin swollen in water and softened
by heating. Diluted to 10% solution in distilled
water; 0.1% sodium benzoate added as a preserva-
,tive. Softening point 25 °C (77 °F).
1% chrome alum added as a stabilizer. Softening
point > 95 (203 °F).
^Figures in brackets indicate literature
references at the end of this paper.
described in the notes may have influenced the
data. It is evident that only RTV 602 and gela-
tin had sufficiently low absorption coefficients
to permit additional attenuation due to scatter-
ing without exceeding the specified overall at-
tenuation coefficient. Both these materials had
disadvantages. The velocity of RTV compounds is
only two thirds of the desired figure. The gela-
tin is liable to drastic changes of dimensions
and properties as water evaporates from the gel.
Consideration was given to accepting a base
material (such as RTV) with a velocity outside
the specified range, and scaling the sizes of
phantom components so that the travel times of
pulsed signals would correspond to those in the
organs being simulated. However, this approach
was judged inherently unacceptable and it was
anticipated that a base material could be found
or formulated with the desired combination of
properties. Also the RTV silicone rubber was be-
lieved to suffer from long term instability of
acoustic properties [2]. In retrospect, it
seems probable that the same precautions to com-
bat instability that are described below for
gelatin-based compounds would be applicable to
RTV silicones in applications where scaling sizes
might be acceptable.
Gelatin gels were considered suitable for sub-
sidiary experiments in which various scattering
materials were dispersed in the gelatin before
molding; the objective was to produce a fine
textured image and the specified attenuation co-
efficient. From both laboratory measurements
and imaging, with the controls on the diagnostic
equipment set for abdominal imaging, it was found
that a 1 percent dispersion of a cellulose fiber
or a 0.05 percent dispersion of glass micro-
spheres in 10 percent gelatin in water produced
the desired results. The cellulose fiber had an
average fiber length of 70 to 140 ym; the glass
microspheres ranged from 10 to 250 \m in diam-
eter.
The next objective was to identify materials
which, when embedded in gelatin, would simulate
the acoustic aspects of abdominal organs. Trial
and evaluation led to a selection of materials
each of which was embedded separately in a gela-
tin block containing scatterers. These blocks
could then be stacked in various sequences and
coupled with mineral oil for trial imaging.
Figure 2 illustrates the various inclusions
imaged at 3.5 MHz. The top block contains the
base scattering material consisting of 1 percent
cellulose fiber in 10 percent gelatin in water.
A gelatin filled natural sponge is placed near
the left edge.
The middle layer contains a soft rubber bulb
filled with saline solution to simulate a large
cyst or fluid filled organ. The bright spot in
the center of the bulb is an artifact caused ap-
parently by a focusing action. The right side
of the center layer contains a natural sponge
filled with the gelatin containing scatterers,
which might be used to simulate metastatic liver
disease.
The lowest level contains a variety of tubing.
The parallel lines display the 1/8 inch wall of a
soft rubber tube filled with clear gelatin. The
upper wall is seen near the junction between the
blocks and the lower wall in the center of the
block. This is offered as an example of the
resolution available at a depth of 14 cm through
324
Fig. 2. Phantom components imaged at 3.5 MHz.
the various phantom components. At the lower
right, some cross sectional images of tubes com-
pare favorably with normal images of the portal
vein or abdominal aorta. The tubes are poly-
olefine shrink tubing 5/16 inch o.d. with 0.02
inch wall and clear vinyl tubing 7/16 inch diam-
eter with 1/16 inch wall. These phantom com-
ponents were scanned at both 2.25 MHz and 3.5
MHz with improvements in resolution at the higher
frequency corresponding to that found in bio-
logical tissue. Figure 3 shows an image at 2.25
MHz with resolution clearly inferior to that of
figure 2.
Fig. 3. Phantom components imaged at 2.25 MHz.
The echo discontinuities and interfaces be-
tween layers should be eliminated when a con-
tinuously molded phantom is produced rather
than the individual blocks shown here.
Gelatin gels were originally considered un-
desirable as the base material because of their
obvious instability under heat and evaporation
of water and their susceptibility to degradation
while nourishing mold organisms. However, the
addition of chrome alum as a thermal stabilizer
and sodium benzoate as a preservative eliminated
two of these undesirable features. The problem
of evaporation of water is not so easily solved
if it is required that the propagation velocity
remain in the range 1540 ± 50 m/s. Two solutions
appear possible: a substantial fraction of the
water can be replaced by hydrophilic compounds
that have at least as high a boiling point as
water and preferably higher. Table 2 shows
some results of this approach; all the velocities
are too high. Alternatively, samples can be
coated to prevent the water from evaporating;
this approach was undertaken.
Table 2.
Velocities at
2.25 MHz, 22
°C.
Composition
Velocity
(m/s)
10% gelatin, 89% water
1% chrome alum
1539 ±
10
10% gelatin,
70% glycerol
20% water
1900
10% gelatin,
67% glycerol
20% water
3% PEG 200
1900
10% gelatin,
45% PEG 600
45% water
1770
10% gelatin,
60% PEOa 200,
20% water
10% sorbitol
1749
10% gelatin,
45% PEG 200
45% water
1 746
10% gelatin,
45% DEG
45% water
1717
10% gelatin,
70% DEGb
20% water
1703
10% gelatin,
60% carbitol
30% v-;ater
1644
^PEG = Polyethylene glycol
DEG = Diethylene glycol
First the base material was slightly modified
by inclusion of 5 percent polyethylene glycol 200
in the formulation, to provide compatibility with
the first of five coatings that were applied to
the test blocks of gelatin. Then, successive
coatings of a butadiene-acrylic rubber, a mixture
of this rubber with paraffin, and two coats of a
mixture of the same rubber with polyvinyl acetate
were applied. By grading the coatings in this
manner, good adhesion and acoustic contact be-
tween coatings was obtained, while providing the
desired properties in succession. The paraffin
prevented water evaporation, and the polyvinyl
acetate provided the tough flexible exterior sur-
face. Water evaporation through this coated
specimen has been reduced to 0.25 percent per day
with the specimen open to air. It is expected
that a practical phantom would be enclosed in a
protective case, within which the humidity can
be somewhat controlled as long as a reservoir of
water is replenished. Consequently, the problem
of water evaporation is considered to be under
control, although further improvements in coating
formulations are under study.
In conclusion, specific combinations of mate-
rials have been identified to make possible the
construction of a phantom object to simulate
many of the acoustic characteristics of human
abdominal tissue. Such a phantom should aid in
the standardization of operating conditions of
ultrasonic diagnostic equipment.
325
Acknowledgments
References
The authors wish to acknowledge the invaluable
assistance of William Mullen in performing acous-
tical measurements, Marie Comas and Irving Illing
in preparing test samples, and Linda McKay, Cheryl
Wilson and Diane Eskelson in scanning the many
test components.
The work reported in this paper was carried out
on behalf of the Picker Corporation.
[1] McSkimin, H. J., Ultrasonic Methods for
Measurement, in Physical Acoustics, W. P.
Mason, ed. , Vol. lA, pp. 271-334 (Van
Nostrand, New York, 1965).
l"2] Private communication at WFUMB Workshop,
August 1976.
326
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer , ed. , National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
TISSUE SIMULATORS FOR DIAGNOSTIC ULTRASOUND
Reginald C. Eggleton and James A. Whitcomb
Ultrasound Research Laboratories of
The Indianapolis Center for Advanced Research, Inc.
Indianapolis, Indiana 46202, U.S.A.
Factors are presented related to the design, construction and use of phantoms to
replace human patients or subjects in the development, testing, clinical training and
promotion of ultrasonic diagnostic equipment. Materials are described which match
the acoustic properties of tissue. Realistic echograms can be obtained by scanning
properly configured phantoms using such materials.
The phantoms may be simple geometric test patterns, sections of human torso or
complete human torsos simulating dynamic cardiovascular and respiratory movements to
evaluate the real-time systems.
Key words: Training phantoms; tissue signature; tissue simulators; ultrasonic
phantoms .
1. Introduction
Tissue simulators or phantoms, as they are
sometimes called, may be utilized in the field
of medical ultrasound for purposes of quantita-
tively and qualitatively evaluating the perfor-
mance of diagnostic or therapeutic equipment.
They may also be used for purposes of training.
Yet another application is basic study of the
interaction of sound with tissues for purposes
of improving existing diagnostic criteria, under-
standing the mechanisms of acoustic image forma-
tion and for devising new diagnostic methods.
The design and construction of realistic
ultrasonic phantoms to substitute for human pa-
tients or subjects during the early training of
technicians and physicians is based upon knowledge
from several diverse fields of study. The anatomy
of the phantom must be correct with respect to the
details visualized by ultrasound. Landmark fea-
tures used by ul trasonographers to orient and
identify the scan planes must be accurately rep-
resented. The acoustic properties of biological
materials important to the appearance of the
image must be catalogued for the tissues that will
be represented in the phantoms. A knowledge of
materials that can be used in the construction of
the phantoms is necessary. It is essential to
know how to control various acoustic properties
such as attenuation, scattering, speed of sound,
impedance, etc. Likewise it is necessary to have
a knowledge of the chemistry of the materials to
select components which will give the phantoms
adequate stability and a useful life.
A knowledge of fabricating techniques required
in the construction of the phantoms is also neces-
sary. The methods for casting and welding of
phantom materials to obtain the shapes and con-
tours necessary to adequately model the human
anatomy are also required. It is also desirable
to have the necessary artistic ability to create
an aesthetically acceptable model which will have
the feel and appearance of the structures it pur-
ports to represent.
2. Materials and Methods
Hydrocolloid gels are ideally suited for con-
struction of ultrasound tissue phantoms because of
the flexibility and control over acoustic param-
eters and the stability with respect to time.
There are two hydrocolloid systems which are par-
ticularly suitable for this application; one which
is reversible, and the other irreversible. A com-
bination of the two allows great flexibility in
fabricating techniques.
Irreversible Hydrocolloid - Alginic acid was
first prepared and analyzed by Stanford in 1886
[1]^ who established it as a weak organic acid
that readily forms salts and bases. Nelson and
Cretcher [2] hydrolyzed alginic acid to obtain the
salt of D-mannuronic acid. X-ray and chemical
evidence suggest that alginic acid is a high molec-
ular weight polymer made up of D-mannuronic acid.
Several different methods indicate that the com-
mercial sodium alginates have a molecular weight
between 3200 and 200,000 and a degree of polymeri-
zation from 0.180 to 0.930. Recent evidence has
been obtained showing that the polymer molecule is
entirely linear, consisting of linked mannuronic
acid units and a!?.(l-^4) linked glutamic acid units
[3-5]. Models constructed from x-ray diffraction
data indicate that the D-mannuronic acid units are
in the C-1 chair configuration, whereas the t-glu-
tamic acid units are in the 1-C configuration [6];
^Figures in brackets indicate literature
references at the end of this paper.
327
thus algin is a copolymer and not simply a mix-
ture of mannuronic acid and gluronan.
1/1
OH
I
c=o
I
c 0
OH
,1
I I
H H
H
I
-C
N H
°M
c
c=o
I
c
H /I
\/\r /\/v
OH/
1/
The copolymers are linked together with an
alternating composition. The statistical descrip-
tion of the structure of algin is best obtained
by assuming that the copolymer is formed accord-
ing to the penultimate-unit theory of addition
copolymerization [7].
Water soluble salts of alginic acid include
those of alkaline metals, ammonia and low molec-
ular weight amines. Sodium alginate is the most
common salt and is readily available. Solutions
can be made in either hot or cold water. Algin-
ates dissolve most easily when sifted into water
and vigorously agitated. High speed stirring and
gradual addition of solid increases the rate of
dissolution. Facilitation of the process can be
achieved by wetting the alginate with ethanol or
glycerol or other water-miscible liquid before it
is added to the water. With a good dispersion
the alginate dissolution should be completed in
a few minutes.
The properties of alginate solutions can be
changed readily to provide either slow, fast or
intermediate types of flow [8]. Algin solutions
are strongly influenced by covalent ions such as
calcium which increase the viscosity and lower the
flow properties, with little change in the acous-
tic properties. Alginates show increasing non-
Newtonian flow characteristics with increased de-
gree of polymerization and substitution of calcium
for sodium [9], and become thixotropic at low
levels of calcium. The effect of this ion in
causing increased viscosity is more pronounced in
alginates with higher D-mannuronic acid content
[10]. The viscosity of sodium alginate solutions
can be depressed by using sodium with potassium
salts. The high viscosity alginate can attain a
viscosity of at least 2000 cps at a concentra-
tion of 1 percent in water, whereas a low viscosi-
ty product will have a viscosity of less than 10
cps at the same concentration as a function of
molecular weight. Algin solutions behave like
other fluids in their dependence of viscosity on
temperature and may decrease approximately 2.5
percent per °C with increasing temperature. The
viscosity of water soluble algin salts changes
only slightly as a function of changes in pH, in
the range of 4 to 10.
The cross-linking created by the substitution
of calcium for sodium renders the alginate in-
soluble. When the insoluble salt is formed by
the reaction of the sodium alginate in solution
with a calcium salt, for example, the calcium ion
may replace the sodium ions in two adjacent mole-
cules to produce a cross-linking between the two
molecules. As the reaction progresses, a cross-
linked molecular complex or polymer network forms.
Such a network constitutes the micelles structure
of the gel. Solutions of sodium alginate are
stable for long periods at room temperature but.
like other polysaccharides, are depolymerized in
the presence of a number of autoxidizable sub-
stances such as phenolic compounds. Sodium algin-
ates tend to retard the growth of many microorgan-
isms or are passive with respect to others. Al-
though the algin is resistant to the attack of
microorganisms, its solutions are subject to some
bacterial action during prolonged storage. Agents
such as Zepharin chloride can be added in small
quantities to further decrease the bacterial
growth. Small amounts of calcium increase the
stability of sodium alginate solutions.
Pure algins in water form gels having excellent
optical clarity. By altering the formulation,
gels can be varied in texture from those which are
soft and exhibit flow characteristics to those
which are tough and elastic. This range of prop-
erties exhibits little effect on the acoustic im-
pedance or acoustic velocity characteristics of
the material. As a result, algin gels are ideal-
ly suited to tissue simulation in that structures
can be molded and the attenuation and scattering
properties can be independently controlled.
In general, the gels are formed by the gradual
and uniform release of either calcium or hydrogen
ions, or a combination of the two throughout the
algin solution. The setting time can be control-
led by the addition of a limited amount of a com-
pound such as phosphate or polyphosphate that is
capable of combining with calcium. Frequently
calcium sulphate and a phosphate buffer are com-
bined to control the gelation rate of the algin-
ate solutions.
The acoustic properties of the gel are deter-
mined largely by the water. Two percent sodium
alginate solution is the usual concentration in
combination with small quantities of salts. The
acoustic velocity using the phase method [13] is
approximately 1550 meters per second at 37 °C;
the density of the 2 percent gel is approximately
1.04 grams per cubic centimeter, thus the acous-
tic impedance falls in the center of the plot of
tissue characteristics (see figure 5). The at-
tenuation for the gel without the scatterers in-
corporated is approximately 0.5 dB cm"^ MHz"i.
Reversible Hydrocolloid - An organic hydro-
philic colloid of polysaccharide can also be uti-
lized for construction of acoustic test objects.
A commonly available form utilizes the sulphuric
ester of a linear polymer of galactose. This
material undergoes a reversible reaction, sol t
gel, under the influence of temperature changes.
In the gel state the binding forces are second-
ary forces, e.g., dipole interactions or van der
Waals forces. When the concentration of the dis-
persed phase in the hydrocolloid is of the proper
amount, the sol can be changed to a semisolid gel
when the temperature is decreased below the criti-
cal point (^^ 110°F). The gelation is complete at
approximately 100°F. In this process the dis-
persed phase agglomerates to form chain fibrils,
referred to as "micelles." The process continues
until a three-dimensional matrix is formed with
much branching and intermeshing of the micelles
to form a "brush heap" structure. The intermesh-
ing and weak binding forces contribute strength
to the structure. The interstices of this struc-
ture are filled with water; the water predominant-
ly contributes to the acoustic properties of the
gel. It will be noted upon warming that the sol
st-ate is again attained but the temperature must
be maintained at 120°F for a period of time be-
328
fore liquefaction is achieved. The temperature
lag between gelation and liquefaction is described
as a hysteresis phenomenon.
In the gel state the structure exhibits both
elastic and viscous properties. If the gel is de-
formed it requires time to recover. Likev/ise, if
the gel loses water, it will tend to absorb water
approaching the initial state. The strength of
the gel is decreased at elevated temperatures,
but can be substantially increased with the addi-
tion of a small amount of borax. The borax ac-
complishes this by increasing the interaction of
the micelle framev^ork. The borax also increases
the viscosity in the sol state. The addition of
absorbers and scatterers also increases the vis-
cosity, as well as adding to the strength of the
gel. A suitable composition would consist of
v/ater (83.5 percent), polysaccharide (14.2 per-
cent), potassium sulphate (2 percent) and borax
(0.2 percent). The potassium sulphate is intro-
duced to minimize the amount of syneresis as well
as imbibition. The introduction of soluble salts
into the gel adjusts the osmotic pressure of the
dispersion medium and thus helps to stabilize the
system. The polysaccharide frequently used is ex-
tracted from seaweed and is a linear polymer of
galactose, shown below.
I
c—
OH/I
/ OH
c
\
c —
I
H
I
c—
OH/1
l/OH
V
i/\ K
Fig. 1. Scattering test objects are constructed
with Dextran spheres dispersed in a re-
versible hydrocolloid (polysaccharide).
The relative concentration ranges from
0.1 to 100 as marked to produce a range
of scattering intensities. To scan the
test object the transducer is placed
against a thin plastic window at the top
of the device.
The acoustic properties of the gel with embed-
ded scatterers and absorbers closely match those
of tissue. The specific acoustic properties can
be adjusted by varying the composition of the gel.
The choice of whether to use the reversible or
irreversible hydrocolloid will be dictated by the
fabrication requirements. For example, an inclu-
sion within a simulated organ could be construct-
ed from an irreversible hydrocolloid and then
later cast in the reversible hydrocolloid so that
the temperature of the embedding material would
not affect the shape of the inclusion. One obvi-
ous advantage of the reversible hydrocolloid dur-
ing the development phase is that the materials
can be re-used, and it is therefore more economi-
cal, but once the design has been established,
this advantage disappears.
The acoustic properties of the reversible poly-
saccharide hydrocolloid gels are: density -
1.044 g/cm3; velocity of sound - 1,540 m/s; at-
tenuation - 0.4 dB cm"i MHz"i. These values are
remarkably close to those measured for the ir-
reversible hydrocolloid and for tissue.
Test objects utilizing this material were con-
structed in which various concentrations of scat-
terers were incorporated as layers. Because there
is no difference in the characteristic acoustic
impedance across interfaces, the only reflection
produced is due to the scattering (no specular
reflection is obtained). Figure 1 shows tv;o test
objects developed to display a wide dynamic range
of backscattering targets. The relative concen-
tration of scatterers is indicated for each level,
and ranges from 0.1 to 100 as marked on the test
object.
Silastic Phantoms - A tissue simulator not
previously described in a publication was submit-
ted to the AIUM Standards Committee in 1970 [12].
This silastic device contained both interfaces of
graded reflectivity and scattering sources. The
test block v;as constructed as indicated in figure
2 using silicone rubber (RTV 3110) and a silicone
TEST BLOCK
97%
915
88%
8S%
100%
0.2
0.4
0.6
0.8
1.0
0.0
100%
COMPOSITION OF TEST SPECIMEN
Composition of mixture
Sound velocity Characteristic
Sil icone
Sil icone
Density
at 37.0 °C
impedance
rubber
fluid
g/cm'
m/s X 10-3
rayl « 10-5
RTV 3110
DE200-20CS
P
C
Z
100 %
0 %
1 .129
0.960
1 .084
97
3
1.124
0.962
1.082
94
6
1 .120
0.964
1.080
91
9
1.115
0.967
1.078
88
12
1.110
0.969
1.076
85
15
1 .105
0.971
1.073
100
0.971
0.975
0.947
Fig. 2. The tissue simulator shown schematically
above was constructed using mixtures of
silastic and fluid to achieve a range of
impedance differences across an interface.
329
Fig. 3. Materials exhibiting a range of acoustic
impedances are achieved by varying the
ratio of a two-component plastic system
where the individual components have dif-
ferent acoustic properties. In this in-
stance, silicone rubber and silicone fluid
are mixed to vary the acoustic impedance
over a I percent range. These components
are then assembled in a test block to
provide interfaces of varying reflectivity
fluid (DC200-20CS) . This mixture provided a
range of acoustic impedances (fig. 3). Suspended
in the silicone fluid were particles whose mean
diameter was 7 pm. These particles represented
a source of acoustic scattering, thus an ultra-
sound B-mode scan of the test block revealed the
interface between silastic blocks of various com-
position and scattering which occurs in the bulk
medium (fig. 4). Panels A, B, C and D show the
test block at four gain levels 40, 30, 20 and
10 dB. An adjustment was then madd to the visuali-
zation system which sacrifices sensitivity for
higher longitudinal resolution, better signal-to-
noise ratio and dynamic range compression, but dis-
plays the weak specular reflectors at the inter-
faces with good "gray scale" and resolution, but
without displaying the scatterers. Panel E shows
the effect of optimizing the system response for
visualizing the interfaces within the test block.
The response curve was adjusted to give an output
of 1 volt for an interface with 1.0 percent dif-
ference in acoustic impedance (minimum detectable
brightness on the display). The echoes from the
scatterers then fell below 0.1 volt level. This
is an example of the way in which the test object
can be used to standardize the equipment perfor-
mance so as to optimize the display of a particu-
lar type of target, organ or condition.
Plastisol Phantoms - Phantoms constructed of
plastisol are more suitable for basic studies on
the interaction of sound and tissue because of
their inherent stability and the precision with
which the characteristic acoustic impedance,
velocity and density can be controlled to achieve
precise acoustic properties. Because of its re-
latively high attenuation it is not ideally suited
for construction of training phantoms where prop-
erties and distances of 15 to 20 cm are required.
By using various plasticizers and by adjust-
ing the resin/plasticizer ratio, it is possible
to modify the characteristic acoustic impedance
of the system throughout the range of soft tissue
structures. The acoustic properties of a system
E
Fig. 4. Ultrasonic scan (at various sensitivities)
of test block shows low intensity specular
reflections at interfaces and scattering
from particles in the bulk material.
using epoxy soya plasticizer are shown in figure
5. Also plotted are a number of tissue charac-
teristics for reference. It will be noted that
it is possible to match the impedance, velocity
or density of most any tissue and that the other
parameters will be nearly equivalent. The at-
tenuation of this plastisol system is higher than
tissue and ranges from about 2 dB cm"i MHz'i to
4 dB cm"i MHz"i for the materials shown in
figure 6.
Analysis of the frequency characteristics is
achieved by making measurements with a range of
transducers operating in a pulsed mode. A Fast
Fourier Transform (FFT) for displaying the fre-
quency spectrum of the pulse transmitted through
the specimen is compared with the spectrum ob-
tained for a pulse transmitted through water
[13]. Figure 7 shows the normalized frequency
response for both the reference and sample wave-
forms. Amplitude comparison between the reference
and sample transformed responses yields the nor-
malized attenuation as a function of frequency,
as shown in figure 8. Additional transducers may
be utilized to extend the frequency range. Phase
comparisons between the reference and sample re-
sponses give the phase velocity, as plotted in
figure 9.
330
1.15 -
1.10
1.05 -
1 .00
0.95
1 .45
Fig. 5. Acoustic properties of a plastisol system
in which the resi n/pl asti ci zer ratio is
adjusted to match the impedance of bio-
logical materials.
n.e 5.
FREOUENC
S.e 7.3
MEGftHERTZ
Fig. 7.
2 3 4
Frequency (MHz)
Fig. 6. The normalized attenuation as a function
of frequency is shown for two plastisol
samples having impedances of 1.52 and 1.63
X 105 Rayl r^espectively. Note that the
former closely matches the impedance of
water so that the reflection losses at the
interface are minimal. In the second
specimen, the reflection losses average
1.4 dB per surface. Inasmuch as most
normal soft tissue has a much lower at-
tenuation rate than that shown for these
two samples, materials of this type would
be utilized to represent highly attenuat-
ing pathological states such as those com-
monly found in breast tumors.
The method for evaluating tissue simula-
tors includes measurement of the fre-
quency response characteristics of the
specimen using a computer-generated
Fourier transform of the pulse propagated
through the specimen. The S201 sample
used here is from the epoxy soya plastisol
family having an acoustic impedance of
about 1.56 X 10^ Rayl. The upper curve,
"S", is the reference response without
the sample. Below it are four sample
responses for four closely spaced points
on the sample. The signal-to-noise ratio
for this run is such that the data in the
3 to 6 MHz range is valid, with the best
accuracy assumed around 4.2 MHz, the fre-
quency of maximum sample signal amplitude.
The sample and reference responses may be
compared both in magnitude and phase.
Magnitude comparison yields the sample
attenuation (see fig. 8) while phase com-
parison gives the sample phase velocity
(see fig. 9) .
Scatterers - The vast majority of echoes ob-
tained by scanning a patient are from scatterers
rather than specular reflectors. These scatter-
ers produce echoes which are not individually re-
solved but add constructively when they are in
phase to produce a spot on the display. To dupli-
cate this phenomenon in tissue phantoms, it is
necessary to embed small scatterers (like the
cells in tissue) whose dimensions are small com-
pared to the wavelength of the sound. Spherical
scatterers adequately represent cells whose geome-
try is essentially spherical, but elongated scat-
terers are required to represent tissues in which
331
51
O
X
2:
2;
o
c.
)
f
•>
*^
I—
, — 4;
i
I
3.0
3.6 4.2
FREQUENCY
4.8 5.4
MEGAHERTZ
6.C
Fig. 8.
The attenuation, normalized in the
customary manner, for the Fast Fourier
Transform responses given in figure 7,
is given here. Only data in the assumed
valid data range of 3 to 6 MHz is pre-
sented; however, the variation in attenua-
tion seen here is still partly due to the
low signal amplitudes near 3 and 6 MHz.
The solid bar at the bottom of the plot
indicates the range of frequencies over
which a least squares curve fit was per-
formed on the attenuation (in dB) vs_.
frequency data. The arrow is on the fre-
quency of maximum amplitude, as described
in figure 7.
the cellular makeup is other than spherical, like
in muscle, for example.
Small plastic microspheres are available in a
variety of material types which can be embedded
in the acoustic medium to produce scattering.
The particles come in a variety of sizes. In
general the particle size should approximate the
dimensions of the scatterers of tissue, if the
frequency dependence of scattering is important.
One of the common sources is the Dextran par-
ticles (Sigma Chemical Company, for example) used
in Sephadex columns. This type of particle tends
to imbibe water, thus the dimension in the dry
state will be different than the dimension in the
aqueous medium. Particles can be sized by putting
the powder through graded sieves. Waag et al .
[14] have used three grades, each of which con-
tains a wide range of sizes. There may be occa-
sions when the range of sizes would be desirable
in simulating tissue. It is characteristic of
tissues to have rather uniform cellular dimensions.
Grade
G50M
G50F
G50SF
Size
86 - 257 ym
34 - 137 ym
17 - 68 ym
The uniformity of unscreened microspheres com-
pared with screened microspheres is shown in
figure 10. A uniform population of microspheres
may be used to study the frequency dependence of
scattering and will simulate tissues having
05
\
o
_J
Ld
>„
>-
>-
r
3.0
Fig. 9.
3.6 4.2
FREQUENCY
4.8 5.4
MEGAHERTZ
The phase velocity of the sample examined
in figures 7 and 8 is presented here.
This phase velocity is defined by the
equation
Cs - (l/C^ + d(fi/df)(l/2Trt)
where C^ is the velocity of the reference
medium, t is the thickness through which
the acoustic pulse travels in the sample,
and d^/df is the slope of the phase angle
vs. frequency data. This latter value is
determined by the least squares technique
over the range of frequencies indicated
by the solid bar at the bottom of the
plot. The arrow is on the frequency of
maximum amplitude, as described in figure
7. For comparison, the group velocity,
that is, the velocity of the pulse con-
taining all frequencies as shown by the
Fourier transform of figure 7, for this
sample is 1.558 m/ms.
spherical cells. The range of frequencies for
the transition from Rayleigh to Bragg scattering
is very broad if the distribution of micro-
spheres as shown in figure lOA is used in a tis-
sue phantom, but is abrupt with scatterers in
figure lOB. This point will be further developed
later (see fig. 14).
Acrylic microspheres are available as a dental
repair material. These microspheres can also be
sifted through graded sieves to obtain uniform
size distribution. In some instances the more
intense scattering provided by the acrylic micro-
spheres is desirable. Further, the dimensional
stability is very much greater, therefore, for
applications in which dimensional stability is
crucial, the acrylic microsphere should be con-
sidered.
Another source of microspherical particles is
Teflon microspheres, which are similar to the
acrylic microspheres in terms of scattering,
strength and dimensional stability. Both the
Teflon and acrylic microspheres may need to be
mixed with a wetting agent, such as alcohol, to
promote suspension. Since these spheres are
denser than water, they tend to settle out faster
than the Dextran.
Fig. 10. The Sephadex column packing spheres are
useful sources of acoustic scatterers
and are available in three size ranges;
however, each of the size ranges has a
spectrum of scatterer sizes. The photo-
graph on the left shows the distribution
of sizes obtained for the medium range.
The photograph on the right shows the
20 ym sizes selected by a screening
process .
Hollow glass microspheres show the greatest
difference in impedance compared with the medium
(fig. IIA). The glass microspheres are lower in
density than water and tend to float.
Between these four spherical scatterers with
their various acoustic properties, it is possible
lj to simulate many different tissue conditions.
? Scatterers are added to the medium in a concentra-
I tion of approximately 1 percent per unit volume.
' For Dextran this gives a backscattering charac-
r teristic equivalent to a highly reflecting tissue,
■j In many cases less than 1 percent is adequate to
'•mimic the desired tissue.
The angular dependence of scattering can reveal
information on the orientation of asymmetrical
I scatterers. It is therefore important to include
ji such sources in the test target that are aligned
in a specific direction. Targets containing such
Fig. 11. Scattering particles consist of cellulose
acetate filaments or plastic spheres.
The acetate fibers are 5 um in diameter
and 0.1 mm in length, and the spheres
shown in the photograph are approximately
10 pm in diameter, but other sizes can
readily be obtained by putting the random
spheres through screens of the appropriate
size mesh. Alignment of the fibers is
achieved by flowing the medium prior to
curing. The number of scatterers per unit
volume, the size of the scatterers, the
difference in acoustic index of refraction
of the scatterers compared with the em-
bedding medium, the form factor, the
diameter/length ratio and the degree of
order are all factors which influence the
statistical distribution of scattering.
scatterers will exhibit angular dependence of scat-
tering more like that of muscle.
Rod-like scatterers can be obtained by using
flocking, which is a cellulose acetate material
(see fig. UB). The flocking is an acetate fila-
ment and is available in various size ranges.
Alignment of the rods can be achieved by causing
the material to flow, then gelling the substrate
before diffusion reorients the rods. The rods
exhibit an angular dependence of scattering when
333
they are all aligned in the same direction.
Flocking is available in a variety of size ranges,
but typical dimensions are 15 micrometers by 1
mi 1 1 imeter.
Attenuators - The attenuation of the hydrocol-
loid materials is low and comparable to water.
The addition of scatterers increases the attenua-
tion, but this may still be below the level needed
to simulate many tissues; therefore, it is neces-
sary to add a substance to further increase the
attenuation without increasing the scattering prop-
erties. A substance to achieve this property would
be any insoluble powder whose particle size is very
small compared with the wavelength of the examining
beam. Numerous powders will fulfill this require-
ment. Among those used is diatomaceous earth, tal-
cum powder, powdered chalk, graphite, lamp black,
etc . We have used diatomaceous earth for this prin-
cipally because of its availability. The powder
is mixed with the scattering spheres and hydrocol-
loid until the desired attenuation has been achiev-
ed. Using this technique the scattering and at-
tenuation can be adjusted independently over some
range of acoustic characteristics.
Fabrication Techniques - Organs can readily be
molded by using natural organs or models of natural
organs. Any of several molding techniques is suit-
able, including gel molds. The positive replica of
the organ is cast in the gel and the mold is par-
tially opened. The replica is extracted by deform-
ing the gel mold.
The surface of the mold may be covered with
glycerin or other suitable parting agent and the
warm sol solution is poured into the mold. Sprues
can be provided to carry away gas as the gas in
the mold is displaced by the sol.
If the organ is to represent a kidney, for in-
stance, it would be desirable to cast the organ in
several stages to build up a structure which simu-
lates the internal architecture of the kidney. In
this case the core is held in place by struts while
the cortex is b^ing cast in the mold. Afterwards,
the struts are removed and the void is filled with
sol, using a syringe. The completed kidney is then
encapsulated in a film to minimize evaporation and
add stability and convenience of handling. The
film can be applied by dipping or by sealing the
material with such as a thin polyethylene over the
surface of the organ and heat sealing the edges.
The very thin polyethylene is easily stretched over
the organ and if the procedure is carefully done,
it is possible to avoid inclusion of air between
the skin and the kidney. The thickness of the skin
will determine the specular reflection from the
surface of the organ.
Specular reflectors can also be generated by
casting smooth interfaces between materials having
different acoustic impedances and scatterers are
introduced by mixing plastic microspheres or rods
in the compound before the heat treating (solva-
tion). The plastic microspheres should be of a
size which is very small compared to the wave-
length and comparable in dimensions to cells in
tissue, i.e., in the order of 10 micrometers in
diameter. The material of the microsphere could
be chosen so that it will not be affected by the
embedding material. In order to avoid introducing
gas bubbles into the mixture it is necessary to
wet the microspheres with a small quantity of
liquid which is miscible with the plasticizer and
which does not interfere with its properties,
such as alcohol .
3. Classification of Tissue Simulators
As noted in the Introduction there are three
major applications for tissue simulators, or phan-
toms: 1) a training device to be used in place of
human subjects; 2) a device for evaluating the
performance of ultrasonic visualization systems;
and 3) a model for studying the basic interactions
between sound and tissue.
Each of the three areas of application imposes
design requirements on the construction of tissue
simulators. The training device should yield
realistic echograms when scanned with conventional
diagnostic equipment so that the trainee can learn
scanning techniques and the interpretation of echo-
grams. The training device should also be fabri-
cated to include both normal and abnormal anatomy
(fig. 12) illustrating the diagnostic criteria.
The phantom should contain all of the necessary
landmarks and reference points used by ultrasono-
graphers in their diagnostic protocol. The dimen-
sions of the training device should hold a 1:1
correspondence with the dimensions of the struc-
tures represented, therefore, the attenuation in
the tissue simulator should correspond closely
with the attenuation experienced in the correspond-
ing human anatomy.
Fig. 12. Simulation of a cross-sectional slice of
the human torso which can be scanned with
an ultrasonic transducer. The resulting
echograms exhibit a 1:1 correspondence to
the same section through the human body.
The training device would be constructed »
to show both "normal" anatomy and simulate!
pathology. '
II
Tissue simulators used to check the performance i,
of ultrasonic diagnostic equipment should include >.
means for examining the sensitivity and resolution c
(both lateral and longitudinal) of the equipment ;
over the dynamic range of signal intensities en- li
countered in clinical applications (fig. 13). It
is desirable that these test objects be stable '
with respect to time so that instabilities in the
diagnostic equipment can be measured. Using the i
simulator as a standard, equipment adjustments can i}
334
10
Fig. 13. A test block incorporating low re-
flectivity specular reflectors,
scatterers and wedges for resolution
measurements are cast in a plastic
block which can be utilized to examine
the performance of visualization
equipment.
be made to optimize the system performance for
displaying certain tissue characteristics, such as
was done in optimizing the display of weak scatter-
ing interfaces in figure 4, panel E.
Tissue simulators used for basic studies will
vary widely in their design as dictated by the ex-
perimental conditions under study. We have been
using simple disc-shaped test objects of unifonri
thickness with known scattering densities and im-
pedance characteristics to examine the angular and
frequency dependence of scattering in tissue. It
is important to control the size of the scatterer.
For scatterers which are very small compared to
the wavelength of sound, scattering is of the Ray-
leigh type and has a fourth pov/er cross section.
In the range of Bragg scattering, there is a power
of two dependence of frequency on the scattering
cross section (fig. 14). This plot is an extension
of data presented by Freese et al . [15] and nicely
illustrates the regions of Rayleigh and Bragg scat-
tering for oil droplets in gelatin and lipid filled
cells in fish muscle.
4. Conclusion
Tissue simulators are necessary to meet various
requirements with respect to the application of
ultrasound to medicine which are not fulfilled by
test targets such as the AIUM 100 mm test object.
Recently there has been an interest in developing
tissue equivalent targets from several sources
and this interest stems from both a requirement
to provide more quantitative diagnostic methods
as well as an interest in reducing unnecessary ex-
posure of patients to ultrasound. Although as yet
no toxic effects have been documented for visuali-
'zation systems used in the customary manner, it is
prudent to minimize unnecessary exposure whenever
possible. Because it is possible to accurately
simulate body structures for ultrasonic scanning,
construction of such phantoms would appear to be
an important current objective.
.01 -
.001 -
.0001
(coho)muscle
with 2% lipid content
\im droplet
backscatter coefficient
Q for oil droplets in
gelatin as a function
of frequency
I I I I I
Fig. 14.
1 10 100
Frequency (MHz)
The plot of the backscatterers ' cross
section vs^. frequency for oil droplets
in gelatin and for lipid droplets in
muscle display both Rayleigh and Bragg
scattering phenomena. The Rayleigh
scattering has a fourth power dependence,
whereas the Bragg scattering has a
second power dependence on frequency.
Acknowl edgments
This work was supported in part by NSF Grant
APR75-15908, a Smith-Kline Fellowship, and by The
Indianapolis Center for Advanced Research. The
computer processing technique was developed by
Francis J. Fry and Narendra T. Sanghvi.
References
[1] Stanford, E. C. C, J. Soc. Chem. Ind. 5,
218 (1886). ~
[2] Nelson, W. L. and Cretcher, L. H., J. Amer.
Chem. Soc. 51_, 1914 (1929).
[3] Atkins, E. D. T., Mackie, W., and Smolko,
E. E., Nature 225, 626 (1970).
[4] Atkins, E. D. T., Mackie, W. , Parker, K. D.,
and Smolko, E. E., Polymer Lett. 9, 311
(1971).
[5] Rees, D. A. and Samuel, J. W. B., J. Chem.
Soc. C, 2295 (1967).
[6] Larsen, B., Painter, T. , Haug, A., and
Smidsr^d, 0., Acta Chem. Scand. 23, 355
(1969).
' Painter, T. , Smidsreid, 0., Larsen, B., and
Haug, A., Acta Chem. Scand. 22, 1637 (1968).
[8] McDowell, R. H. , J. Soc. Chem. Ind. (London),
Monograph No. 24, 19 (1966).
335
[9] Hirst, E. L., Percival, E., and Wold, J. K.,
Chem. Ind. (London), 257 (1963).
[10] Smidsrjid, 0. and Haug, A., Acta Chem. Scand.
19, 329 (1965).
[11] Skinner, E. W. and Phillips, R. W., eds..
The Science of Dental Materials, pp. 101-
135 (W. B. Saunders Co. , Philadelphia, 1967).
[12] Eggleton, R. C, Fabrication of Ultrasonic
Test Specimen for Evaluating the Performance
of Echo-Ranging Equipment, submitted to AIUM
Standards Committee (1970).
[13] Franklin, T. D., Jr., Sanghvi, N. T., Fry,
F. J., Egenes, K. M., and Weyman, A. E.,
Ultrasonic tissue characterization studies
of ischemic and infarcted myocardium,
presented at 2nd International Symposium on
Ultrasonic Tissue Characterization, Session
5; June 13-15, 1977, Gaithersburg, Maryland.
[14] Waag, R. C. et al . , personal communication
(advanced copy of manuscript).
[15] Freese, M. L. and Hamid, M. A. K., Lipid
content determination in whole fish using
ultrasonic pulse backscatter, in 1974 Ultra-
sonics Symposium Proceedings, pp. 69-76,
IEEE Cat. No. 74 CH0896-1SU.
336
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer , ed., National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
TISSUE EQUIVALENT TEST OBJECTS FOR COMPARISON OF ULTRASOUND
TRANSMISSION TOMOGRAPHY BY RECONSTRUCTION WITH PULSE
ECHO ULTRASOUND IMAGING
Paul L. Carson and Leonard Shabason
Department of Radiology
University of Colorado Medical Center
Denver, Colorado 80020, U.S.A.
Donald E. Dick
Department of Physical Medicine
University of Colorado Medical Center
Denver, Colorado 80020, U.S.A.
and
Wi 1 1 iam dayman
Alderson Research Laboratories, Inc.
Stamford, Connecticut 06903, U.S.A.
Tissue equivalent materials have been investigated for evaluation and comparison of
pulse echo ultrasound imaging and ultrasound transmission tomography by reconstruction
(UTTR). Investigations have centered primarily on various urethane polymers and 3M
Reston Brand Flotation Pad material. Attenuation coefficients of the urethane polymers
still are somewhat too high, and thus initial test objects or "phantoms" have been con-
structed from the flotation pad material. One phantom chosen to simulate several char-
acteristics of human breast tissue consists of an annulus of unaltered flotation pad
material surrounding a center region in which scattering polystyrene microspheres are
embedded. Contrasting material such as polyethylene rods can be inserted between the
inner and outer areas of the phantom. The UTTR technique clearly delineates poly-
ethelene rods in the phantom as small as .6 mm in diameter. The scattering at 2.2 to
3.5 MHz nominal frequency exhibits a pulse echo appearance, similar to that of liver
tissue, and causes only approximately a 5 percent increase in the attenuation coeffi-
cient of the pure flotation pad material .
Key words: Computed tomography-ultrasonic; diagnosis-ultrasonic; tissue equivalent
test objects and phantoms; ultrasound imaging; ultrasonic tissue char-
acterization.
1. Introduction
Among the various properties of tissue that
can be detected with ultrasound, attenuation and
velocity have attracted a great deal of atten-
tion. This interest has been intensified by
evidence that malignant and benign breast tumors
can be distinguished from each other and from
normal tissue by significant differences in at-
tenuation of ultrasound by these different tis-
sues [1]^. The goal of the main project lead-
ing to this study has been to develop a system
for the quantitative estimation of the spatial
'Figures in brackets indicate literature
references at the end of this paper.
distribution of attenuation and velocity co-
efficients in living tissue. This instrumenta-
tion development has been accompanied by the
development of various tissue-equivalent test
objects or phantoms to help ensure accurate and
rel iable results.
Simultaneous measurement of velocity and at-
tenuation can be performed with direct ap-
proaches to ultrasound computerized tomography
(CT), although many questions exist as to the
accuracy of these measurements. To study the
potential of quantitative and visual diagnosis
with the technique, a system for performing
ultrasound transmission tomography by recon-
struction (UTTR) has been designed and con-
structed [2]. The attenuation image shown
here was obtained with this new instrumentation,
337
using principles delineated in reference [3].
Basically, opposed transmitting and receiving
transducers are scanned on either side of the
imaged object in the translate-rotate motion
of early x-ray CT scanners. The amplitude of
the rectified and integrated pulse waveform
was linearized and used as the input variable
in a convolution reconstruction program. The
complete UTTR system soon will include the
capability for simultaneous generation of
pulse-echo images as well as CT scans of veloc-
ity, and attenuation. The system has enough
flexibility to serve as a general purpose re-
search instrument, but the initial investiga-
tions will focus on breast imaging. This em-
phasis was chosen because of the high incidence
of breast cancer in women coupled with the need
for more reliable, non-invasive techniques to
detect and characterize breast lesions. Another
factor is the existence of preliminary encourag-
ing results obtained by others with ultrasound
velocity CT images [4].
To provide a controlled comparison of the
various imaging approaches, tissue equivalent
objects have been developed. This development
has followed three approaches which have been
pursued concurrently:
1. Development and identification of
stable plastics which are equivalent
to tissue in terms of physical density,
sound propagation velocity, and atten-
uation as a function of frequency.
2. Simulation of pulse echo scattering
using small reflectors embedded in
available plastics.
3. Molding the available tissue equivalent
plastics and scattering materials into
phantoms to allow evaluation and com-
parison of alternative imaging approaches.
2. Developments
To date, most of the investigation on tissue
equivalent plastic has been directed to various
urethane polymers such as TDI polyether urethane
prepolymer with high molecular weight polyol
chain-extenders, and to the 3M Company material
used in Reston Flotation Pads. The urethane
polymers can be made with densities and ultra-
sound propagation velocities in the soft tissue
range. Shown in figure la is an example of at-
tenuation measurements on one of these polymers,
which has a density and ultrasound propagation
velocity at 27 ^C of 1.1 x 10^ kg/m3 and 1544
m/s, respectively. Work is progressing on reduc-
ing the ultrasound attenuation coefficient in
these materials to that of various soft tissues.
Advantages of this class of polymer are that the
materials are rigid enough to support test ob-
jects in known locations, properties can be
varied, and the materials are stable in time.
The 3M Flotation Pad material exhibits ultrasound
interaction characteristics similar to that of
fat except for a somewhat higher attenuation co-
efficient as shown in figure lb. The amplitude
attenuation coefficient in SI units is related
to the attenuation coefficient in dB/cm by
)j(m-M = 11.513 • a(dB/cm). Also, p/f (s/m) =
1.1513 X 10-- 5 • a/f (dB/cm-MHz). The speed of
ultrasound propagation in the flotation pad ma-
terial at 27 "^C is 1461 m/s and its density is
9 X 10^ kg/m^. Figure 1 was obtained using
broadband transmitters with working frequencies
of 2.2 and 4 MHz with diameters of 19 and 13 mm,
respectively. A 0.25 mm diameter miniature
hydrophone was employed as a receiver in the far
field of the transmitter, and attenuation of
gated sine waves was measured by interposing and
removing samples of the materials under study.
Polystyrene microspheres in the range of 15
to 300 )im in diameter have been embedded in
various plastics, and their pulse echo imaging
characteristics have been studied in densities
of 0.2 to 2.0 spheres per um^. At these den-
sities, 80 ym diameter polystyrene microspheres
scatter more weakly than normal liver, as de-
termined with commercially available pulse echo
imaging systems at 2.2 - 5 MHz. Consequently,
polystyrene microspheres of 175 ym diameter, in
densities of approximately 0.3 spheres per mm^,
have been chosen for initial simulation of soft
tissue scattering.
The phantom designed for comparisons of the
various ultrasound imaging techniques is dia-
f(MHz)
Fig. 1. Semilogarithmic plots of attenuation coefficient as a function of frequency at two temperatures
(a) urethane polymer; (b) 3M flotation pad material.
338
Fig. 2. Top view and side view of the Tissue
Equivalent Test Object. The central
cylinder contains isotropic scatterers.
Rods and tubes may be inserted between
the central cylinder and the outer an-
nulus. The three nylon bolts shown in
the drawing have been replaced by 0.13
mm diameter nylon monofilament.
grammed in figure 2. An annulus of 10 cm outside
diameter consisting of unaltered flotation pad
material surrounds a 4 cm diameter of cylinder of
the flotation pad material containing the scatter-
ing microspheres. This 5 cm thick cylinder is
sandwiched between two acrylic plates for geometri-
cal stability. Holes in the acrylic plate located
conveniently at the interface between the outer
annulus and the central cylinder allow rods and
liquid-filled cylinders of various attenuating
materials to be inserted between the scattering and
nonscattering materials.
3. Examples of Use
Pulse echo images of the phantom obtained with
a commercial ultrasound scanner are shown in
figure 3. Polyethylene rods of 1.9, 2.7 and 3.3
mm diameter are displayed from left to right at
the edge of the inner cylinder. In this image,
the polyethylene rods are in the focal zone of
the 3.5 MHz transducer and the shadowing of the
scattering material by the rods is visualized
easily. Figure 4 is an UTTR image of attenuation
in the same phantom with rods of 3.3, 2.7, 1.9
and 0.6 mm diameter. As may be seen, each of the
four rods was delineated clearly. Phase cancella-
tion at the receiver due to the increased speed
of sound in the rods may have contributed to
their detectabi 1 ity in both the attenuation image
and the pulse echo shadowing of discrete scat-
terers. The same effect should contribute to
the detection of highly attenuating lesions in
the body.
Attenuation coefficients for the flotation pad
material obtained in this reconstruction are
a = 2.88 ± 0.01 dB/cm or p = 33.2 ± 0.2 m-i. The
quoted errors are the standard error of the mean
of several hundred pixels in the reconstructed
image. These attenuation coefficients correspond
to a/f = 0.82 dB/cm-MHz at the 3.5 MHz working
frequency employed for the reconstruction. If
the frequency averaged attenuation coefficient
of a/f = 0.9 dB/cm-MHz at 21 °C is accepted from
figure lb as the true value for the flotation pad
material, then the effective frequency of the
Fig. 3. Pulse echo image of the tissue equiva-
lent phantom using a conventional scan-
ner with a 3.5 MHz transducer focused
at the top side of the central cylinder
containing scattering particles.
Shadowing of the scatterers by 3.3, 2.7,
and 1.9 mm polyethylene rods is apparent.
Fig. 4. Attenuation image of the phantom by
ultrasound transmission tomography by
reconstruction. The polyethylene rods
visible in the image range in diameter
from 0.6 to 3.3 mm.
339
broadband ultrasound pulses used to create the
UTTR image may be taken to be 3.2 MHz. By com-
parison, the mean attenuation coefficient of the
flotation pad material containing scatterers is
3.02 ± 0.02 dB/cm. This measured value in the
central scattering region may be slightly low due
to beam hardening effects. Nevertheless, the
measured difference in the attenuation coefficient
of the material with and without scatterers is
only 5 percent, and the real difference probably
is not much greater. The greater attenuation in
the scattering material is visualized easily in
UTTR images displayed at high contrast.
The maximum reconstructed attenuation coeffi-
cients at the centers of the 3.3, 2.7, i.9 and
0.6 mm diameter polyethylene rods are 20, 12, 12
and 6 dB/cm, respectively. Clearly, these ap-
parent attenuation coefficients for the smaller
rods are reduced by resolution effects. Bulk
measurements of attenuation coefficients of the
polyethylene rods were not obtained, because the
rods were made by heating and stretching larger,
low density polyethylene rods. This process
probably would affect the density of the poly-
ethylene, and consequently the ultrasound attenua-
tion coefficient.
Pulse echo imaging of large samples of the
flotation pad material containing polystyrene
microspheres has indicated that the flotation
pad material at 21 °C attenuates ultrasound some-
what more rapidly than normal liver tissue in
vivo. It was assumed, therefore, that the flota-
tion pad phantom would be a rigorous test of UTTR
and pulse echo systems for breast imaging. Ini-
tial measurements on the female breast in vivo,
indicate that there are relatively large quanti-
ties of tissues in many young, female breasts
which are much more attenuating than anticipated.
These materials, presumably dense, fibrous tis-
sues, often may be aligned parallel to the skin
surface and thus provide a long, attenuating
path for transmission imaging in coronal sections
of the breast. Attenuation coefficients averag-
ing 15 dB/cm in these highly attenuating tissues
have been measured with the 3.5 MHz pulses. It
should be noted, however, that much of the at-
tenuation obtained in the UTTR system is due to
phase cancellation and refraction at the inter-
face between tissues or phantom materials with
relatively low and relatively high speeds of
ultrasound propagation.
The breast phantom developed to date will be
useful for comparing UTTR and pulse echo imaging
under the best conditions, but much more complex
phantoms will be required to simulate many im-
portant features of normal and abnormal breasts.
Acknowl edgments
This investigation was supported in part by
grants APR 76-05944 from the National Science
Foundation and 5T32 CA 9073 awarded by the
National Cancer Institute, Department of Health,
Education, and Welfare.
We gratefully acknowledge the contributions
of the University of Colorado Medical Center
personnel: Will Carter, Tom Oughton, Mary Dick,
Elliott Bayly, Gary Thieme, and William R.
Hendee, as well as the contributions of Paul E.
Hansen, of the 3M Company, who provided sample
materials and useful information.
References
[1] Calderon, C, Vilkomerson, D., Mezrich, R.,
Etzold, K. F., Kingsley, B., and Haskin, M. ,
Differences in the attenuation of ultra-
sound by normal, benign and malignant
breast tissue, J. Clin. Ultras. 4, 249-254
(1976).
[2] Dick, D. E., Bay, H. P., and Carson, P. L.,
Hardware Design of an Ultrasound CT Scanner,
Proceedings , Rocky Mountain Biomedical En-
gineering Symposium, published April 1977
by the Instrument Society of America.
[3] Carson, P. L., Oughton, T. V., Hendee, W. R.,
and Ahuja, A. S., Imaging soft tissue
through bone with ultrasound transmission
tomography by reconstruction. Medical Physics
4, 302-309 (1977).
[4] Greenleaf, J. F., Johnson, S. A., Samayoa ,
W. P., Duck, F. A., and Wood, E. H. , Alge-
braic Reconstruction of Spatial Distributions
of Acoustic Velocities in Tissue from their
Time of Flight Profiles, in Acoustical
Holography, N. Booth, ed.. Vol. 6, pp. 71-90
(Plenum Press, New York, 1975).
340
APPENDIX
I
341
Reprinted from Ultrasonic Tissue Characterization II, M. Linzer , ed.. National Bureau
of Standards, Spec. Publ. 525 (U.S. Government Printing Office, Washington, D.C., 1979).
APPENDIX
DATA OF THE VELOCITY AND ATTENUATION OF ULTRASOUND IN
MAMMALIAN TISSUES - A SURVEY
R. J. Parry and R. C. Chivers
Physics Department
University of Surrey
Guildford, England
A compilation of the reported values of velocity and attenuation of ultrasound in
mammalian tissues is presented to give a clear picture of the state of knowledge and
enable it to be assessed. In the twenty years since the last such compilation, the
progress in estimating the relative contributions of the animal species, tissue con-
dition, temperature and frequency at which measurements were made and the method of
measurement to the observed variation has been small. It is hoped that this compila-
tion will both be of practical use and also encourage the establishment of a stronger
body of fundamental information for the application of ultrasound in medicine and
biology.
Keywords: Attenuation; mammalian tissues; ultrasound; velocity.
1. Introduction
With the increasingly recognized importance of
fundamental work on the interaction of ultrasound
with tissue there is a need for ready access to
currently available data of the acoustic parameters
of mammalian tissues. There are summaries of ve-
locity and attenuation data [1-3]^ but all lack
some of the information needed to make a critical
assessment of the present state of knowledge. In
view of this we have attempted here to produce a
survey which is both accurate - not always a fea-
ture of previous surveys - and useful in providing
such information.
The grouping of the data is primarily by type
of tissue and mammalian species and subsequently
by tissue condition and temperature. Tissue condi-
tion in vitro is categorized as fresh, fixed or
refrigerated (though for comparisons to be proper-
ly valid more precise details of the history of the
tissues must be known).
It is assumed that the measurements were made at
sufficiently low intensity such that non-linear ef-
fects were not significant. However, only Dussik
et a1 . [4] and Fry and Fry [5] report values for
ultrasonic intensity. This may arise from the
present state of the literature on the measurement
of intensity [6].
, The quoted attention coefficient, a, is defined
by A = Aoe""^, where A is the amplitude of a plane
i, wave of initial amplitude Ag after travelling a
;' distance x in the tissue. Assuming a realistic
ll plane wave, or that any geometrical effect on the
I amplitude has been removed, this represents the
^Figures in brackets indicate literature
references at the end of this paper.
sum of the energy absorbed within the tissue and
the energy scattered by inhomogeneities in the
tissue [7]. It must also be assumed that the in-
tensities used were low enough for there to be no
irreversible changes caused in the tissue and thus
that a is not dependent on the intensity. To con-
vert the values quoted to decibels they should be
multiplied by 8.686.
The tables include the number of measurements,
N, and some indication of the spread of values
about the quoted mean: where possible the stand-
ard deviation (SO) or standard error (SE) of the
set of measurements, other'wise the range of values
or some unspecified "variation".
As a guide to what may be expected of the mea-
surements reported, it was thought useful to in-
dicate the measurement technique used. For the
velocity measurements these have been divided in-
to pulse transmission or reflection (pulse), in-
terferometric and sing-around techniques and for
attenuation into broadband, using pulses contain-
ing only a few cycles of vibration, or narrow band,
with 20 or more cycles. Results obtained with
broadband pulses [10-12] are usually quoted as
being at one particular frequency without any de-
tail of how this frequency was selected. For pref-
erence the frequency specified should be that of
the maximum amplitude in the frequency spectrum of
the pulse. Recently the whole frequency spectrum
of a broadband pulse has been used to measure at-
tenuation [7,13] and there is the possibility of
using swept frequency techniques [60]. Further
consideration o1^ measurement methods is out of
place in the present context and the reader is
referred to the review article by McSkimin [8] and
the work of Wehr [9,14] for discussion of the
various techniques and the relative errors in-
volved.
343
The information collected in the tables below
is that relating to whole tissues only. Thus much
data on blood is excluded as it is concerned with
the constituents rather than whole blood (see Hus-
sey [3] for a summary), and in other reports the
"whole blood" is a dilution with plasma of centri-
fuged red blood cells [57,59]. Nor was it felt ap-
propriate to include all the data from isolated
references on the variation of parameters with
other features of the tissues, for example the
variation of velocity in fat with water content
[15,16] and the work on liver homogenates [17,18].
For such information the reader is referred to the
original literature.
Summary
The result of a survey such as this should per-
haps be some set of values of parameters or curves
that may be taken to be characteristic of the vari-
ous tissues considered. However, considering the
data presented here, all that may be concluded is
that at the present time it is virtually impossible
to separate any variation between different types
of tissue from those variations due to the many
other factors involved, such as the measurement
technique, temperature, condition and treatment of
the tissue, the species from which the tissue came
and its pathological condition. To produce any
"characteristic" values many assumptions would have
to be made, the validity of which cannot be assessed
from the information available. Thus all that the
present authors have set out to do is to present
the data with the relevant information, where pos-
sible, in order that each worker may draw his own
conclusions about average values and their' limita-
tions.
The table A, for velocity data and B, for at-
tenuation data contain this information; blank (--)
indicates that the appropriate information is not
stated in the original paper. To examine the fre-
quency and temperature dependence of velocity, the
(a priori unwarranted) assumption is made that the
results for different species and conditions may
be directly compared and the measurements for two
temperature ranges are then represented in figures
1 and 2. The ranges chosen were 23 + 3 °C and
36 ± 1 °C, the former being rather wide in order
to encompass the large body of measurements made at
20 °C and 26 °C. It is left to the reader to opine
about the existence of velocity dispersion or to
assess temperature dependence of velocity from
significant differences between figures 1 and 2 -
a task made difficult by the lack of data at 36 ±
1 °C (although some data in the tables does not ap-
pear in these figures as it was averaged over a wide
frequency range [10]). The remaining figures (3-9)
show plots of a/f against f for various tissues -
no account being taken of species, condition or
temperature. This presentation was chosen as it
readily permits assessment of the oft-quoted linear
relation between a and f. Comments on the tempera-
ture dependence of a are limited by the consistent
disinclination of authors to specify this parameter
when reporting their results and their tendency to
average over wide temperature ranges.
In conclusion, it is only possible to echo Gold-
man and Hueter's statement of 20 years ago, "No
critical discussion is indicated at this time, but
it is anticipated that the accumulation of further
data on the basis of more precise and extensive
measurements should permit important generaliza-
tions on the acoustic characteristics of living
matter." [1], and in doing so express the hope that
future reports in the literature will include suf-
ficient information about tissue condition and mea-
surement technique to enable valid conclusions about
their contribution to the spread of parameter values
to be assessed and thus permit the "important
generalizations on the acoustic characteristics of
1 iving matter. "
Table. Compilation of (A) velocity and (B) attenuation of various
tissues under various conditions.
Species
Condi ti on
Velocity of ultrasound in various tissues under various conditions
N Technique
Temp .
(°C)
f
(MHz)
(m/s)
Variation
(m/s)
Precision Source
it)
Normal Blood
man
heparini sed
22
4
2
1565
125
12
sing-around
1
19
22.
6
1570
24
10
23
2
1549
6
10.8
10
24
2
1556
4
4.7
11
Normal Fat
man, orbit
fresh
20
6 -*
14
1582
SD = 20.4
65
pulse
1
5
10
37
1462
SD = 23.7
16
man, breast
fresh/dry
25
2
0
1470
pul se
0
2
20
man, breast
refrigerated
24
1
8
1465
SE = 2
interferometric
0
5
16
man
24
1
8
1476
SE = 2
2752
interferometric
0
5
16
pig
fresh
24
1
8
1444
SE = 2
58
interferometric
0
5
16
pig
fresh
37
.5^7
1440
21
cow
fresh
24
1
8
1465
SE = 3
21
i nterferometri c
0
5
16
horse
fresh
24
1
8
1443
SE = 4
56
i nterferometri c
0
5
16
344
Species Condition Temp. f c Variation N Technique Precision Source
{°C) (MHz) (m/s) (m/s) (%)
Normal Liver
man
—
24
1.8
1585
SE = 2
--
i nterferometri c
0.
5
16
pig
fresh
24
1.8
1587
SE = 3
--
i nterferometri c
0.
5
16
pig
fresh
25
2.5
1553
SD = 15.5
7
sing-around
0
5
22
cow
fresh
24
1.8
1590
SE = 2
--
i nterferometri c
0
5
16
cow
fresh
24
1.8
1578
SE = 3
i nterferometri c
0
5
16
horse
fresh
1 Q
i . O
iooU
i nter f erome trie
u
C
ID
dog
fresh
26
4
& 12
1581
SD = 16.8
8
i nterferometri c
0
2
23
rabbi t
fresh
24
1.8
1599
SE = 1
i nterferometri c
0
5
16
rabbi t
fresh
26
4
& 12
1575
SD = 9.4
13
i nterferometri c
0
2
23
guinea pig
fresh
24
1.8
1575
SE = 3
i nterferometri c
0
5
16
Abnormal Liver
man, healthy
refrigerated
24
1.8
1570
SE = 4
18
interferometric
0.
5
16
rabbit, bled
f res h
9/1
1 . o
1607
SE = 2
1 n eerie rome ir i c
u
D
1 f\
guinea pig,
bled
fresh
1 o
1 .8
1589
Norma'
SE = 2
Kidney
interferometri c
0.
5
16
pig
fresh
24
1.8
1560
SE = 4
—
interferometric
0
5
16
pig
fresh
25
2.5
1558
SD = 15.6
5
si ng-around
0
5
22
cow
fresh
1.8
1568
SE = 3
i nterferometri c
U
c
3
i D
cow
fresh
24
1.8
1572
SE = 3
i nterferometri c
0
5
16
horse
fresh
24
1.8
1558
SE = 3
interferometric
0
5
16
doq
fresh
26
4
& 12
1559
2
interferometric
0
2
23
rabbit
5
1566 1560 - 1571
2
24
Norma
Spleen
pig
fresh
24
1.8
1578
SE = 3
interferometric
0
5
16
pig
fresh
c . b
1515
SD = 15.2
4
SI ng-around
u
c
D
09
LL
cow
fresh
24
1.8
1577
SE = 2
interferometric
0
5
16
cow
fresh
24
1.8
1578
SE = 3
—
interferometric
0
5
16
horse
fresh
24
1.8
1595
SE = 3
interferometric
0
5
16
dog
fresh
26
4
& 12
1570
2
interferometric
0
2
23
Normal Lung
dog
fresh
35.0 ±
.5
0.98
650
0
2
ref lecti on
coefficient^
25
dog
fresh
35.0 ±
.5
1
5
658
812
1180
—
reflection
coef f i ci ent^
less than
10 ms'i
52
doa
0. 39
1
300
580
pulse
55
Abnormal Lung
dog, pneu-
monitic
fresh
35
0.98
340
0
1
reflection
coefficient^
25
Normal Connective Tissue
man, breast
fresh/dry
25
2.0
1545
Norma'
Muscle
pul se
0.
2
20
Skeletal muscle: parallel
to muscle
f i bres
dog
fresh
26
4
& 12
1592
SD = 5.0
7
interferometric
0
2
23
rabbit
fresh
26
4
& 12
1603
SD ^ 7.9
13
interferometric
0
2
23
345
Species Condition Ternp. f c Variation N Technique Precision Source
(°C) (MHz) (m/s) (m/s) (%)
Skeletal muscle: perpendicular to muscle fibres
man, external
fresh
20
6
14
1612
SD
12.5
83
pul se
1
5
10
eye
37
1631
SD
_
15.3
13
man, breast
fresh/dry
25
2
0
1545
pul se
0
2
20
dog
fresh
26
4
&
12
1576
SD
-
9.6
8
i nterferonietri c
0
2
23
rabbi t
fresh
26
4
&
12
1587
SD
5.5
18
i nter f erometri c
0
2
23
Skeletal muscle
: direction
unspecified
man, pectoral
ref ri gerated
24
1
8
1568
SE
=
5.5
--
interferometric
0
5
16
man
--
24
1
8
1585
SE
=
6
interferometric
0
5
16
pig
fresh
24
1
8
1580
SE
3
interferometric
0
5
16
cow
fresh
C H
1
8
1581
SE
4
i nterf erometri c
0
5
16
cow
fresh
9/1
1
1
8
1580
SE
4
interferometric
0
5
16
cow
refrigerated
25
2
5
1580
SD
=
16
12
sing-around
0
5
22
horse
fresh
24
1.
8
1598
SE
4
_ _
interferometric
0
5
16
Cardiac muscle
pig
fresh
24
1
8
1587
SE
3
i nterf erometri c
0
5
16
cow
f re s h
24
1
8
1584
SE
3
"i rThprfpyrirnptri c
0
5
16
cow
fresh
24
1
8
1570
SE
2
interferometric
n
u
c
J
1 F.
10
horse
fresh
24
1
8
1584
SE
3
* ~
interferometric
0
5
16
dog
fresh
26
4
&
12
1572
3
interferometric
0
2
23
Normal Nervous Tissues
man, brain
fresh
24
5
1524
pul se
0
1
26"
(term foetus)
24
1521
il
1540
man, brain
1525
27
(foetus )
man, brain
24
1
8
1564
SE
4
—
interferometric
0
5
16
pig, brain
fresh
24
1
8
1565
SE
2
i nterferometric
0
5
16
pig, brain
fresh
25
2
5
1506
SD
15
7
interferometric
0
5
22
cow, brain
fresh
24
1
8
1560
SE
2
i nterf erometri c
0
5
16
cow, brain
fresh
9/1
1
8
1569
SE
3
inter icruiMcLi \l
n
J
1 u
horse, brain
fresh
24
1
8
1560
SE
2
IMLtri IcrUllltrLi IL
I u
dog, brain
fresh
25
2
5
1515
SD
15
5
si ng-around
U
r
b
99
LC
cat, brain
1 i vi ng/f resh
24
4
2
1557
pul se
- "
11
37
1570
60
1574
rabbit, brain
--
5
1508
1380 -
1570
3
24
man, optic
fresh
20
6
->•
14
1644
SD
25.4
30
pulse
1
5
10
nerve
37
1615
SD
3.1
13
man, cerebro-
fresh
21.8
2
1499
30
11
sing-around
0
1
19
spinal fluid
24.4
1515
45
9
25.0
1509
5
7
.5
11
Abnormal Nervous
Tissues
man,
fixed 3 h
19.0
2
1524
2
6
.1
20
sing-around
0.
1
19
meningioma
48 h
19.8
1524
5
7
.5
20
man ,
19.7
2
1557
31
20
sing-around
0.
1
19
meningioma
1546
15
20
(5 sections
1569
31
21
of 1 tumor)
1548
31
14
1569
39
14
man, glioma
fresh
22.2
2
1529
1
9
.1
20
sing-around
0.
1
19
man, glioma
fixed
22.3
2
1500
45
17
si ng--around
0.
1
19
346
Species Condition Temp. f c Variation N Technique Precision Source
(°C) (MHz) (m/s) (m/s) (%)
man ,
fresh
27.5
2
1545
4 6.2
41
sing-around
0.1
19
astrocytoma
19
man ,
T 1 xeci
2
1517
121
27
<^ 1 n n - ri mi in H
0.1
astrocytoma
man ,
f i xed
20.0
2
1501
45
18
sing-around
0.1
19
ependymoma
Normal Ophthalmic Tissues
man, lens
in vivo
body
--
1585
± 74b
53
—
—
28
man, lens
fresh
34.1
4
1641
SD = 16^
7
i nterferometric
__
29
(mean )
man, lens
fresh
37
4
i nterferometri c
0.6
30
(autopsy)
1641
0 SD = 1.3
35
(operative)
1638
4 SD = 3.0
12
(autopsy + operative)
1640
5 SD = 1.2
47
pig, lens
fresh
23
4
i OD J
= fie
interferometric
29
1 n
JU
1673
SD = 56
10
35
1677
SD = 3^
10
cow, lens
fresh
22
4
iODU
1 nterferometri c
0.5
31
cow, lens
refrigerated
26-31
5
ID iO
iOUU iOOU
i nterferometri c
1
12
rabbit lens
5
i J H-U
OU DO
R
J
24
man, vitreous
in vivo
body
7
lb44
± ii
c 0
bi
28
man, vitreous
fresh
35.2(mean) 4
1530
SD = 5'=
10
interferometric
--
29
man, vitreous
fresh
3/
4
i n ter f erome tr i c
0.6
30'-
(autopsy)
1532
SD = 0.6
35
(operative)
1531.
7 SD = 0.9
14
(autopsy + operative)
T C 0 0
A c n — n c
4 oU - (J . b
4y
pig, vitreous
fresh
23
4
1510
SD = 3^
10
interferometric
--
29
30
1522
SD = 3^
9
35
1531
SD = 3e
10
cow, vitreous
fresh
22
4
1495
interferometric
0.5
31
cow, vitreous
refrigerated
26-31
5
1516
1490 - 1544^
interferometric
1
12
rabbit.
5
1472
SD = 16
6
—
24
vitreous
vitreous + hyal uronidase
1494
SD = 11
6
cow, aqueous
fresh
22
4
1495
interferometric
0.5
31
cow, aqueous
refrigerated
26-31
5
1497
1481 - 1525^
12
man, cornea/
in vivo
body
1502
± 45b
53
28
aqueous
cow, cornea
fresh
22
4
1550
interferometric
C.5
31
cow, sclera
fresh
22
4
1630
interferometric
0.5
31
Abnormal Ophthalmic Tissues
man, lens-
fresh
35.8
4
1643
1
interferometric
—
29
cataractous
1688
1
Normal Bone
Long bones: a
long axis
cow, phalanx
fresh
5
4030
SD = 110
252
pulse
SD 3
32
cow, phalanx
dried
5
4360
SD = 170
120
pul se
SD 3
32
cow, femur
dried
5
4060
SD = 40
144
pul se
SD -x. 3
32
guinea pig.
.1
3158
2870 - 35419
40
pul se
—
33
femur
Long bones: across axis
cow, phalanx
fresh
5
3160
SD = 170
252
pulse
SD -v- 3
32
cow, phalanx
dried
5
3270
SD = 160
120
pulse
SD 3
32
cow, femur
dried
5
3420
SD = 340
144
pul se
SD -x. 3
32
347
Species Condition Temp.
(°C)
f c Variation
(MHz) (m/s) (m/s)
Technique
Precision Source
(%)
Long bones: direction unspecified
man, cortical in vivo body
bone
5.0 3406
fresh
fixed
dog, tibia
horse
Skull bone
man
man
outer layer
diploe layer
inner layer
guinea pig, fresh
completely healed
broken femur
partially healed
broken femur
refrigerated 22
body
& 5 -- 3210
3700
0.8 3360
._h
2920 SD
3198 SD
3098 SD
SD = 126
36209
18
160
93
220
23
9
5
6
man, pre-
menopause
in vTvo
body
Abnormal Bone
0.1 2968 2788 - 33719 15
2551 2442 - 27249 I6
Normal Breast Tissue
2.0 1510 1450 - 1570J ^ 110 pulse
pulse
pulse
pulse
continuous wave
reflection
pulse
pulse
post-menopause
1468 1430 - 1520J
Abnormal Breast Tissue
'x- 40
brain
17 weeks
(2 samples)
28 weeks
(2 samples)
40 weeks
(2 samples)
24
24
37
37
24
24
37
37
24
24
37
37
1490
1495
1520
1523
1498
1502
1528
1529
1521
1524
1540
1540
6
0.2
34
35
36
37
38 i
33
20^
man, carcinoma
refrigerated
24
1
8
1573
SE = 7
interferometric
0
5
16
20^
man, carcinoma
in vivo
body
2
0
1478
SD = 28
17
pulse
0
2
man, fibro-
cystic
in vivo
body
2
0
1531
SD = 23
28
pul se
0
2
20''
man, fibro-
adenosis
in vivo
body
2
0
1529
SD = 30
6
pul se
0
2
20^
man, fibro
adenoma
in vivo
body
2
0
1529
SD = 21
6
pul se
0
2
20^
Obstetric Tissues
man, cervix-
pregnant
nonpregnant
in vivo
body
5
1525
1633
SD = 1.63
SD = 2.86
108
29
pulse
006
39
man, amniotic
fluid
25
5
1510
40
man, milk
30
2
1540
pul se
0
2
20
man, foetal
fresh
5
pul se
0
1
26^
348
Species
Attenuation of ultrasound in various tissues under various conditions.
Condition Temp. f a Variation N Technique Precision Source
Temp.
(°C)
f a. Variation
(MHz) (m-i) (m'M
Norma
D 1 ooci
58
1.09
--
--
--
—
i!
0
1.73
1.
8
3.22
3.
0
6. 33
4'.
8
12!l
DO
1.15
1.
0
1.78
1.
8
3.45
3.
0
6.91
4.
8
12.7
Normal Fat
5.
gHl
230
SD = 22
9 - 13
broadband
—
6.
2
220
SD = 15
7.
5
330
SD = 24
7.
7
250
SD = 18
8.
4
320
SD = 26
8.
8
360
SD = 15
9.
8
430
SD = 20
9.
8
450
SD = 24
10
46
500
SD = 30
13.
9
D/U
c n - OQ
0
180P
--
bjJcL-Lr Ulll ailu lj'-> ' J
4.
4-
Ul a ur UClUUaMU
1 7
360
n 1 1 1 c
|J U 1 b tr
1
10
typi cally
at least
spectrum
5m" 1
2
22
SD = 3
6 per
a 1 1 a 1 y J 1 J u 1
3
35
point:
a broadband
4
47
pul se'^
5
63
1 point
6
84
per 100
7
115
kHz
8
5
i
6.9
SD = 2.3
narrow band
. 25
3
18
SD = 2.3
5
26
SD = 8.1
1
r
D
6.9
--
2
8
21
4
40
6
56
7
75
87
4.6
10%
1
7
8.7
3
4
16
Norma'
Liver
1
1
0 1
20.3
900
broadband
1
13.8
typically
at least spectrum analysis
5m" 1
2
27.6
SD = 6
6 per
of a broadband
3
43
point:
pulse"
A
't
56
5
69
1 point
6
81
per 100
7
93
kHz
1
5
16
2
4
18
4
5
40
1
5
15
2
4
18
4
5
38
dog fresh
(citrated)
30
40
man, orbit
fresh
37
man, orbit
man
in vivo
f i xed
18 ± 2
man
man/cow
pig
fresh
fresh
cow
man
man
(melted)
in VIVO
f i xed
37
20 - 35
body
18 ± 2
cow
cow
fresh
fresh
20 - 35
41
10
61
42
4
21
43
44
7
45
46
349
Species
Condi tion
Temp.
(°C)
f
(MHz)
(m-i)
Variation
(m-i)
Technique Precision Source
cow
fresh
25
7.7
+ 10% 17
cow
cow
ground (± ^ 0.5)
refrigerated
fixed
20 - 35
.35
.575
.6
.7
.3
.87
13
20
29
54
71
104
4
6
10
9
11
9.
10%
47
43
1 7
1 4
97
cow
2
12
—
-.r
54
12
narrow band^
Normal
Kidney
man/mouse
fresh/frozen/ 22 - 31
96
1000
SD = 86
7
--
+ 25%
48
fixed
222
5000
SD = 760
4
cow
fresh 20 - 35
1.5
19
—
—
—
45
2.4
26
4.5
51
cow
fresh
1 t;
i . D
1 Q
46
9 A
CO
4.5
53
cow
fixed
.25
5
—
—
47
.3
4
. 35
5
'a
6
.5
7
.6
5
.7
9
.8
7
pig
2
18
__r
—
54
16
narrow band^
Norma'
Spleen
man
fixed 18+2
1
3.5
typically
at
least
spectrum
5 m- 1
7
2
13
SD = 4.6
6 per
analysis
3
26
point:
of a broadband
4
40
1 point
pul se"
5
56
per
100
6
74
kHz
7
96
Normal Lung
man
fresh
1
350
narrow band
0.25 dB
4
dog
fresh 35.0 ± .5
.98
470.0
10.0
2
25
dog
fresh 35.0 ± .5
1
430
410 - 470^'
52
2.25
590
560 - 640J
5
1160
L040 - 1210J
dog
fresh 27 ± 2
2.4
440
SE = 160
broadband
0.1 dB
56t
5.0
900
SE = 140
7.4
1000
SE = 130
dog, pneu-
moni tic
fresh
35.0 ± .5
Abnormal Lung
350
25
350
Species
Condition
Temp .
f
(MHz)
(m-i)
Variation
(m-i)
Technique Precision Source
Normal Connective Tissues
man, skin fresh
—
1
40
SD =
14
3
85
SD =
14
5
106
SD =
25
man/cow, tendon fresh
1
54
SD =
12
across grain
3
125
SD =
23
5
195
SD =
28
along grain
1
58
SD =
21
man/cow fresh
—
1
38
SD =
10
articular
3
81
SD =
9
capsule
5
130
SD =
62
ma n / rnui "fyf^ Q h
■Mail/ uuw 1 1 Coll
I
58
SD =
Ij
cartilage
3
144
SD =
23
5
220
SD =
62
cow, elastic fresh
1
73
SD =
2
tendon, across
3
188
SD =
2
grain
5
286
SD =
53
along grain
1
41
SD =
8
3
137
SD =
3
5
235
SD =
66
cow, rectum fresh
1
6.9
SD =
1.7
wall
3
18
SD =
2
5
28
SD =
5
Norma'
Muscle
^l/ala^al miicr'To* naval lol
■jKc 1 c Lo 1 IIIUbv.,lc> parallel
to muscle fibres
man fresh
37
5.
gm
180
SD =
18
6.
2
210
SD =
18
7.
5
0/1 n
SD =
15
7.
7
220
SD =
16
8.
4
310
SD =
24
8.
8
320
SD =
25
9.
8
410
SD =
22
9.
8
400
SD =
24
10.
46
440
SD =
20
13
9
590
SD =
17
man/cow fresh
1
16
SD =
3
3
48
SD =
9
5
71
SD =
17
cow
20 -
35
3
9.0
87
18.0
1.
7
25.4
3
4
62.1
Skeletal muscle: perpendicular to muscle fibres
man fresh
37
5
gm
120
SD =
16
6
2
170
SD =
21
7
5
200
SD =
22
7
7
130
SD =
18
8
4
220
SD =
33
8
8
280
SD =
35
9
8
290
SD =
20
9
8
300
SD =
23
10
46
370
SD =
24
13
9
480
SD =
18
man/cow fresh
1
8
SD =
1
3
30
SD =
6
5
40
SD =
1
cow
20 -
35
.3
7.5
.87
5.5
3
.4
26.5
narrow band
0.25 dB 4
9-13 broadband
10
narrow band
0.25 dB 4
10% 43
9-13 broadband
10
narrow band
0.25 dB 4
10% 43
351
Species Condition Temp. f a Variation N Technique Precision Source
(°C) (MHz) (m-i (m-i)
Skeletal
muscle: direction unspecified
man
__
8
9.5
--
--
42
cow
2
--
21 -
1159
r
54
—
21 -
249
narrow band^
Cardiac
muscle
cow
fresh 20 - 35
1
5
30
-
—
45
2
4
45
4
5
80
cow
T rs s n —
1
i
5
30
_.
46
0
L.
4
45
A
5
80
cow
fixed
O
C
D
35
5
4
8
575
10
6
10
7
14
8
14
dog
fresh
?
c
20
SD =
10
12 per spectrum analysis --
62
4
32
SD =
14
point: of a broadband
6
50
SD =
14
1 point pulsei^
8
66
SD =
14
per 500
10
92
SD =
20
kHz
Tongue:
parallel to fibres
cow
iresn cu - jd
1
1
5
23
45
2
4
32
4
5
65
Tongue:
perpendicular to fibres
cow
fresh 20 - 35
1
5
45
45
2
4
65
Tongue: direction unspecified
cow fresh 1.5 20
2.4 32
4.5 62
dog, infarcted fresh -- 2
cardiac 4
muscle 6
8
10
Abnormal Muscle
0
SE =
10
7 per
spectrum analysis
20
SE =
10
point:
of a broadband
60
SE =
15
1 point
pulse"
115
SE =
15
per 500
190
SE =
10
kHz
man, brain
man, brain
pig, (whole
brain)
(whi te
matter)
pig, brain
cow, brain
fixed
fresh
fresh
fixed
fresh
20 - 35
20 - 35
.3
.87
1.7
3.4
1
2.5
2.5
.35
.6
.8
.87
1.7
3.4
Nerve Tissues
8.5
14.0
18.0
36.5
17
40
18
3
4
5
8.5
14.0
33.5
10%
36 - 51
16 - 21
12
6
10%
43
49
50
47
43
352
Species Condition Temp. f a Variation N Technique Precision Source
(°C) (MHz) (m-i) (m-i)
dog, brain in vivo
body
97
5.4
--
1%
41
LaLi Urdin vivu
37
2
28
50
— hrriri Hhfln d —
U 1 <J U 1 1 <-!
11
rat, spinal in vivo
30-31.5
98
10
8
12
7 narrow band
5
cord
mouse, spinal in vivo
2 ± .1
1
1.7
—
53
cord
10 ± . 1
5.0
28 ± . 1
9.5
Parallel to nerve fibres
man, optic fresh
37
5.
gm
290
SO
14
9 - 13 broadband ■ —
10
nerve
6.
2
280
SO
16
7.
5
430
SD
22
7
7
400
SD
47
8.
4
600
SD
=
38
8.
8
500
SD
38
g
8
620
SD
38
9
8
670
SD
=
36
10
46
600
SD
=
21
13
9
740
SD
28
man, medulla
20-35
1
7
14
10%
43
oblongata
3
4
34
cow, sciatic
20-35
3
4
40
nerve
Perpendicular to nerve fibres
man, optic fresh
37
5
gm
210
SD
=
20
9-13 broadband
10
nerve
6
2
240
SD
=
20
7
5
310
SD
=
28
7
7
230
SD
=
22
8
4
370
SD
=
43
8
8
350
SD
=
29
y
Q
. O
Jdu
c n
oU
9
8
410
SD
33
10
46
500
SD
=
29
13
.9
630
SD
33
man, medulla
20-35
1
7
21
- - 10%
43
obi ongata
3
4
46
cow, sciatic
20-35
3
4
55
10%
43
nerve
Normal Ophthalmic Tissues
man ,
lens
10°
92P
spectrum analysis —
13
i
+
of a broadband
17
156
pul se'^
cow ,
lens
fresh
28
3.25
64
6
4
12
cow,
lens
22
10
230
58
13.5
310
cow.
vitreous
22
6
6
58
10
12
18
20
30
29
cow.
aqueous/
fresh
25-28
30
33
9
large broadband 15%
12
vitreous
Normal
Bone
man ,
skull
fresh
.3
23
7
broadband
51
.6
52
16
.8
92
28
1.2
170
51
1.6
320
96
1.8
430
130
353
Species Condition Temp. f a Variation N Technique Precision Source
(°C) (MHz) (m-i) (m-i)
2
3
25
5
530
780
160
230
man, skull
fresh/fixed
8
150
_ _
__
37
man, skull
outer
diploe
inner
fixed
_.
.h
1919
1616
2460
SD = 474
SD = 293
SD = 1250
24
9
5
6
25%
38^
man, skull
1
150
49
man/cow
fresh
1
144
narrow band
0.25
dB
4
dog, tibia
refrigerated 22
3
c
0
150
220
—
—
—
35
horse
q
1
2
4
43
86
5
250
920
460 - 580
36
Obstetric Tissues
cow, uterus
fresh
1
19.6
SD = 2
narrow band
0.25
dB
4
The velocity was calculated from measurements of
the reflection coefficient made with a thermo-
couple probe in a standing wave field.
'^The authors do not indicate what this figure
represents .
"Measurements were made on samples from eyes with
a variety of tumors although no pathological
condition was seen in the components of the eye.
"^Reference 30 contains information on the variation
of velocity with time after death.
'Data for the calculation of the standard deviation
were taken from the graphs in the reference.
''calculation of the standard deviation was not
possible since each figure quoted by the author
is a mean of 16 to 25 measurements.
""The authors do not give individual measurements
but only quote this range.
The authors do not specify the frequency used,
except that the pulse used "contained substantial
components in the range 3 to 4 MHz."
'The "average value" for bone quoted by these
authors must be disregarded because of the manner
in which it was obtained.
^Only the range can be obtained from the data which
are represented graphically.
^This includes data from U.S.A. and Australia, and
a study comparing measurements with radiographic
findings.
Data used to calculate the mean and standard
deviation were those marked "a" in tables II to V
of reference 20 (i .e. those breasts diagnosed by
pathology studies of excised tissue).
'"These figures refer to the dominant frequency in
the acoustic pulse from a reflector in water, but
the authors comment that this may change by as
much as 12 percent.
"This method gives the attenuation as a continuous
function of frequency, but the authors did not
include diffraction corrections necessary for
accurate results.
°Frequency resolution 300 kHz.
^The attenuation coefficient was assumed to be
linear with frequency.
"^This author also gives some indication of the tem-
perature variation of the attenuation coefficient.
r
Pulse technique using piezoelectric receiver.
^Continuous wave technique using radiation balance
receiver.
^These authors also give information on the vari-
ation of the attenuation coefficient with the
degree of inflation of the lung.
'^Reference 26 contains further information on the
variation of velocity in foetal brain with the
gestational age of the foetus.
354
oSpleen oCardiac muscle
o Fat Oliver "Kidney •Skeletal muscle
1700p I 1 r 1 I 1 ( 1 I
I .
8
1650
1600
1550
1500
1450-
1400L I I I I I I I 1 I I I I I L
o
• o
■ Sclera
o Brain o Vitreous o Cornea
• CSF •Lens •Aqueous
1 I
n r
J 1 I L
_ I I I I
1231231 231231 2345634534 5
Frequency (MHz)
Fig. 1. Frequency dependence of the velocity of ultrasound in various normal tissues at 23 ± 3 °C.
Spleen Cardiac muscle
Fat Liver Kidney Skeletal muscle
1700p I 1 I 1 I 1 I 1 I
1650
1600
^ 1550
1500
1450
1400
Sclera
o Brain oVitreous Cornea
CSF •Lens Aqueous
1 I I I 1
I L
I I I I \ I I L
I I I I
J I L
( L
1 231 2 31 231 231 2 3 4 5634 534 5
Frequency (MHz)
Fig. 2. Frequency dependence of the velocity of ultrasound in various normal tissues at 36 ± 1 °C.
355
100
■ Buschmann et al . [10]
— Chivers and Hill [7]
0 Pohlman [42]
1 Dussik et al . [4]
a Schwan et al . [21 ]
• Colombati and Petralia [43]
Q Goldman and Hueter [1]
(pooled data)
■ /
Fig. 3. Attenuation of ultrasound in normal fat.
0.1
Frequency (MHz)
100
100
♦ Mountford and Wells [44]
— Chivers and Hill [7]
V Hueter [45]
T Hueter and Pohlman [46]
o Pauly and Schwan [17]
iEsche [47]
• Colombati and Petralia [43]
C3 Goldman and Hueter [1]
(pooled data)
Fig. 4. Attenuation of ultrasound in normal liver.
1|_
0.1
lOOr
100
Frequency (MHz)
Kidney: i. Kessler [48]
7 Hueter [45]
T Hueter and Pohlman [46]
i Esche [47]
Spleen: —Chivers and Hill [7]
Uterus: • Dussik et al . [4]
Fig.
5. Attenuation of ultrasound in normal abdominal
organ tissue (other than liver).
0.1
Frequency (MHz)
100
356
100
Fig. 6. Attenuation of ultrasound in normal muscle.
O tfn
Skeletal: along • Buschmann et al . [10]
across a
along ♦Dussik et al . [4]
across o
along •Colombati and Petralia [43]
across o
oPohlman [42]
Cardiac: Allueter [45]
AEsche [47]
Tongue: along THueter [45]
across v
CjGoldman and Hueter [1]
(pooled data)
Frequency (MHz)
100
(Man)
(Cow)
oColombati and Petralia [43]
(White matter) v
iBallantine et al . [49]
'Hueter and Bolt [50]
□ Vosioka et al . [41]
^Esche [47]
I Robinson and Lei e [11]
QGoldman and Hueter [1]
(pooled data)
Fig, 7. Attenuation of ultrasound in normal brain.
iL_
0.1
lOOr
Frequency (MHz)
□ 0
o a
100
Medulla oblongata:
Spinal cord:
along DBuschmann et al. [10]
across ■
along oColombati and Petralia [43]
across ♦
along OColombati and Petralia [43]
across •
vFry and Fry [5]
Fig. 8. Attenuation of ultrasound in normal nervous
tissues (other than brain).
0.1
100
Frequency (MHz)
357
1000
vDussik et al. [4]
oAdler and Cook [35]
□ Kishimoto [36]
Skull bone
• Hueter [51]
■ Theismann and Plander [37]
iBallantine et al . [49]
100
Fig. 9. Attenuation of ultrasound in normal bone.
10 100
Frequency (MHz)
References
[1] Goldman, D. E. and Hueter, T. F., Tabular
data of the velocity and absorption of high-
frequency sound in mammalian tissues, J.
Acoust. Soc. Am. 28, 35-37 (1956).
Errata: J. Acoust. Soc. Am. 29, 655 (1957).
[2] Wells, P. N. T., Absorption and dispersion
of ultrasound in biological tissue. Ultra-
sound in Med. Biol. 1, 369-376 (1975y^
[3] Hussey, M. , Diagnostic ultrasound (Blackie,
Glasgow, 1975).
[4] Dussik, K. T., Fritch, D. J., Kyriazidou, M. ,
and Sears, R. S., Measurements of articular
tissues with ultrasound. Am. J. Phys. Med.
3^, 160-165 (1958).
[5] Fry, W. J. and Fry, R. B., Temperature
changes produced in tissue during ultrasonic
irradiation, J. Acoust. Soc. Am. 25 , 6-11
(1953).
[6] Zieniuk, J. and Chivers, R. C, Measurement
of ultrasonic exposure with radiation force
and thermal methods, Ultrasonics 14, 161-
172 (1976).
[7] Chivers, R. C. and Hill, C. R., Ultrasonic
attenuation in human tissue, Ultrasound in
Med. Biol. 2, 25-29 (1975).
[8] McSkimin, H. J., Ultrasonic Methods for
Measuring the Mechanical Properties of
Liquids and Solids in Physical Acoustics,
W. P. Mason, ed. , Vol. 1, Part A, pp. 271-
334 (Academic Press, New York, 1964).
[9] Wehr, J. S., Classification and discussion
of interference methods for measuring the
propagation velocity of ultrasonic waves, in
Proceedings of the Fourth International
Conference on Acoustics, Copenhagen, Paper
K15 (1962).
[10] Buschmann, W., Voss, M. , and Kemmerling, S.,
Acoustic properties of normal human orbit
tissues, Ophthal. Res. 1, 354-364 (1970).
[11] Robinson, T. C. and Lele, P. P., An analysis
of lesion development in the brain and in
plastics by high-intensity focussed ultra-
sound at low-megahertz frequencies, J.
Acoust. Soc. Am. 5^, 1333-1351 (1972T.
[12] Begui, Z. E. , Acoustic properties of the
refractive media of the eye, J. Acoust. Soc.
Am. 2i, 365-368 (1954).
[13] Lizzi, F., Katz, L. , St. Louis, L. , and Cole-
man, D. J., Applications of spectral analysis
in medical ultrasonography. Ultrasonics 14,
77-80 (1976).
[14] Wehr, J. S., Error analysis as a factor in
the choice of the proper method of measur-
ing the attenuation and the reflection coef-
ficient of ultrasonic waves in solids, in
Proceedings of the Conference on Acoustics
of Solid Media (PWN Warsaw, 1964), pp. 239-
245.
[15] Frucht, A. H. , Die Geschwindigkeit des Ultra-
schalls in menschlichen und tierischen Ge-
weben, Naturwi ssenschaf ten 39, 491-492
(1952).
[16] Frucht, A. H., Die Schallgeschwindigkeit in
menschlichen und tierischen Geweben, Ze i t .
Ges. Exptl. Med. 120, 526-557 (1953).
[17] Pauly, H. and Schwan, H. P., Mechanism of
absorption of ultrasound in liver tissue,
J. Acoust. Soc. Am. 50, 692-699 (1971).
[18] Danckwerts, H. J., Discrete relaxation
processes as a model of the absorption of
ultrasound in liver homogenate, J. Acoust.
Soc. Am. 55_, 1098-1099 (1974).
358
[19] Van Venrooij, G. E. P. M. , Measurement of
ultrasound velocity in human tissue. Ultra-
sonics 9. 240-242 (1971).
[20] Kossoff, G., Fry, E. K. , and Jellins, J.,
Average velocity of ultrasound in the human
female breast, J. Acoust. Soc. Am. 53,
1730-1736 (1973y:
[21] Schwan, H. P., Carstensen, E. L. , and Li, K. ,
Heating of fat-muscle layers by electromag-
netic and ultrasonic diathermy. Trans. Am.
Inst. Elec. Engrs. 1, 72, 483-488 (1953).
[22] Ludwig, G. D. , The velocity of sound through
tissues and the acoustic impedance of tissues,
J. Acoust. Soc. Am. 22, 862-866 (1950).
[23] Goldman, D. E. and Richards, J. R., The mea-
surement of high frequency sound velocity in
mammalian soft tissue, J. Acoust. Soc. Am.
26, 981-984 (1954).
[24] Nakajima, A., Nishi, S., Amano, K. , Uesugi ,
T. , and Wakai , T. , Measurement of the axial
length of the eye by ultrasonic device, (in
Japanese), J. Clin. Ophth. 14, 1594-1599
(1960).
[25] Dunn, F. and Fry, W. J., Ultrasonic absorp-
tion and reflexion by lung tissue, Phys. Med.
Biol. 5, 401-410 (1961).
[26] Wladimiroff, J. W., Craft, I. L., and Talbert,
D. G. , In vitro measurements of sound velocity
in human fetal brain tissue. Ultrasound in
Med. Biol. U 377-382 (1975).
[27] Willocks, J., Donald, I., Duggan, T. C, and
Day, N. , Foetal cephalometry by ultrasound,
J. Obstet. Gynaec. Brit. Cwlth. 71., 11-20
(1964).
[28] Nakajima, A., Kimura, T., and Yamazaki, M.,
Applications of Ultrasound in Biometry of
the Eye in Ultrasonics in Ophthalmology,
R. E. Goldberg and L. K. Sarin, eds., pp.
124-144 (W. B. Saunders, London and New York,
1967).
[29] Jansson, F. and Sundmark, E., Determination
of the velocity of ultrasound in ocular tis-
sues at different temperatures. Acta Ophthal .
39, 899-910 (1961).
[30] Jansson, F. and Kock, E., Determination of
the velocity of ultrasound in the human lens
and vitreous. Acta Ophtal . 40, 420-433 (1962).
[31] Oksala, A. and Lehtinen, A., Measurement of
the velocity of sound in some parts of the
eye. Acta Ophtal. 36, 633-639 (1958).
[32] Lang, S. B., Ultrasonic method for measuring
elastic coefficients of bone and results on
fresh and dried bovine bones, I.E.E.E. Trans.
Bio-Med. Eng. BME-17, 101-105 (1970).
[33] Floriani, L. P., Debevoise, N. T., and Hyatt,
G. W. , Mechanical properties of healing bone
by the use of ultrasound, Surg. Forum 18, 468-
470 (1967).
[34] Craven, D. A., Constantini, M. A., Greenfield,
M. A., and Stern, R., Measurement of the
velocity of ultrasound in human cortical bone
and its potential clinical importance. An
in vivo preliminary study. Invest. Radiol. 8,
72-77 (1973).
[35] Adler, L. and Cook, K. V., Ultrasonic para-
meters of freshly frozen dog tibia, J. Acoust.
Soc. Am. 58, 1107-1108 (1975).
[36] Kishimoto, T. , Ultrasonic absorption in bones,
Acustica 8, 179-180 (1958).
[37] Theismann, H. and Pfander, F. , Uber die
DurchlHssigkeit des Knochens fUr Ultra-
schall, Strahlentherapie 80, 607-610 (1949).
[38] Martin, B. and McElhaney, J. H., The
acoustic properties of human skull bone,
J. Biomed. Mater. Res. 5, 325-333 (1971).
[39] Bakke, T. and Gytre, T. , Ultrasonic measure-
ment of the sound velocity in the pregnant
and the non-pregnant cervix uteri, Scand.
J. Clin. Lab. Invest 33, 341-346 (1974).
[40] Zana, R. and Lang, J., Interaction of ultra-
sound with amniotic liquid, J. Acoust. Soc.
Am. 55, 391 (1974).
[41] Yosioka, K. , Ohmura, A., Hasegawa, T. , and
Oka, M., Absorption coefficient of ultra-
sound in soft tissues and their biological
conditions, in Proceedings of the 6th Inter-
national Congress on Acoustics, Tokyo, 1968,
pp. M5-M8.
[42] Pohlman, R., Uber die Absorption des Ultra-
schalls im menschlichen Gewebe und ihre Ab-
hclngigkeit von der Frequenz, Physi kal i sche
Zeit. 40, 159-161 (1939).
[43] Colombati, S. and Petralia S. , Assorbimento
di ultrasuoni in tessuti animal i. La Ricerca
Scientifica 20, 71-78 (1950).
[44] Mountford, R. A. and Wells, P. N. T. , Ultra-
sonic liver scanning: the quantitative
analysis of the normal A-scan, Phys. Med.
Biol . 17_, 14-25 (1972).
[45] Hueter, T. F., Messung der Ul traschal 1 absorp-
tion in tierischen Geweben und ihre Abh^ngig-
keit von der Frequenz, Naturwi ssenschaften
35, 285-287 (1948).
[46] Hueter, T. F. and Pohlman, R. , Eine optische
Methode zur Bestimmung der Ultraschal labsorp-
tion in undurchsichtigen , schal Iwei chen
Medien, Z. Angew. Phys. I, 405-411 (1949).
[47] Esche, R., Untersuchungen zur Ultraschall-
absorption in tierischen Geweben und Kunst-
stoffen, Akust. Beih. 2, 71-74 (1952).
[48] Kessler, L. W., VHF ultrasonic attenuation
in mammalian tissue, J. Acoust. Soc. Am. 53,
1759-1760 (1973).
[49] Ballantine, H. T., Hueter, T. F., and Bolt,
R. H., On the use of ultrasound for tumour
359
detection, J. Acoust. Soc. Am. 26, 581 (1954).
[50] Hueter, T. F. and Bolt, R. H., An ultrasonic
method for outlining the cerebral ventricles,
J. Acoust. Soc. Am. 23, 160-167 (1951).
[51] Hueter, T. F., Messung der Ul traschal 1 absorp-
tion im menschlichen SchSdel knochen und ihre
AbhSngigkeit von der Frequenz, Naturwissen-
schaften 39, 21-22 (1952).
[52] I Dunn, F., Attenuation and speed of ultrasound
in lung, J. Acoust. Soc. Am. 56, 1638-1639
(1974) .
[53] Dunn, F., Temperature and amplitude depend-
ence of acoustic absorption in tissue, J^.
Acoust. Soc. Am. 34, 1545-1547 (1962).
[54] Marcus, P. W. and Carstensen, E. L., Problems
with absorption measurements of inhomogeneous
solids, J. Acoust. Soc. Am. 58, 1334-1335
(1975) .
[55] Reid, J. M., (1965), reported in F. Dunn,
P. D. Edwards and W. J. Fry, Absorption and
Dispersion of Ultrasound in Biological Media,
H. P. Schwan ed., pp. 205-332 (McGraw-Hill
Book Co., New York, 1969).
[56] Bauld, T. J. and Schwan, H. P., Attenuation
and reflection of ultrasound in canine lung
tissue, J. Acoust. Soc. Am. 56, 1630-1637
(1974).
[57] Carstensen, E. L., Li, K. , and Schwan, H. P.,
Determination of the acoustic properties of
blood and its components, J. Acoust. Soc. Am.
25, 286-289 (1953).
[58] Filipczynski , L., Etienne, J., Lypacewicz, G.,
and Salkowski, J., Visualizing internal struc-
tures of the eye by means of ultrasonics, Proc.
Vibration Problems 8, 357-368 (1967).
[59] Urick, R. J., A sound velocity measurement for
determining the compressibility of finely di-
vided substances, J. Appl. Phys. 18^, 983-987
(1947).
[60] Heyser, R. C. and Le Croissette, D. H., A new
ultrasonic imaging technique using time delay
spectrometry, Ultrasound in Med. Biol. U
119-131 (1975).
[61] Lizzi, F. L. and Laviola, M. A., Power spectra
measurements of ultrasonic backscatter from
ocular tissues, in Proceedings of 1975 I.E.E.E.
Ultrasonics Symposium, pp. 29-32 (1975).
[62] Yuhas, D. E., Mimbs, J. W., Miller, J. G.,
Weiss, A. N., and Sobel , B. E., Changes in
ultrasonic attenuation indicative of regional
myocardial infarction, in Proceedings of
World Federation of Ultrasound in Medicine
and Biology, San Francisco, 1976, pp. 1883-
1894 (1976).
360
SUBJECT INDEX
Absorption, 19, 29, 43, 153
Acoustoel ectri c effect, 63
Amino acids, 19
Anisotropy, 73, 189
Attenuation, 19, 37, 43, 63, 73, 81,
85, 93, 101, 121, 125, 153, 157
197, 203, 209, 247, 255, 281, 287,
343
Backscattering, 153, 157, 165, 247,
275, 327
Beam displacement, 203
Beam distortion, 203
Blood, 165, 173
Bone, 19, 179, 189, 197
Brain, 81, 203, 209
Breast, 13-16, 85, 93, 173, 221, 255
Cell detachment, 317
Cepstrum, 287
Collagen, 19, 73, 179, 189
Color-coded B-scan, 255, 261
Computed tomography, 227, 235, 247,
255, 337
Convolution, 287
Correlation analysis, 275
De-convolution, 287
Dispersion, 81 , 189, 197
Doppler, 173, 227, 235
Dynamic autocorrelation analysis, 275
Elasticity, 73, 189
Estimation theory, 125
Eye, 19, 111
Filtering, 299-300
Hamming window, 281
Heart, 43, 73, 153, 267
Homomorphic processing, 287
Hydroxyapatite, 179, 189
Hyperthermia, 57
Image analysis, 255, 303
Imaging,
annular array, 255
brain (transkul 1 ) , 235
Doppler (moving media), 235
image reconstruction (see computed
tomography)
synthetic focus, 235
Impedance, 81
Inhomogenei ty thermal losses, 37
Interferogram, 73
Kidney, 303
Lag windows, 281
Lipids, 157
Liver, 125, 153, 157, 287, 303
Lung, 19, 135, (see scattering)
Microscopy, myocardium, 73
Myocardium, 63, 73, 153, 267
Opto-acoustic visualization, 255
Pattern recognition, 297, 303
Phantoms, 179, 323, 327, 337
Phase compensation, 209, 235
Polypeptides, 19
Probability density functions, 125
Prostate, 297
Proteins, 19, 157
Pulse compression, 255
Relaxation phenomenon, 29
Scattering, 37, 111 , 135, 143, 153,
157, 281 (see also backscattering)
Sensitivity enhancement, 255
Signal averaging, 255
Skull , 197, 203, 209
Spectral analysis, 43, 111, 125, 143,
261 , 267, 281 , 287
Speed, (see velocity)
Temperature dependence of ultrasonic
properties, 57, 63, 203, 227, 235
Temporal changes, 267, 275
Thermal "lesions", 203
Thermal wave, 37
Thermodynamics, 189
Tissue death, 317
Tomography, (see computed tomography)
Transducer calibration, 255
Velocity, 19, 43, 53, 57, 63, 73, 81,
179, 189, 197, 343
Visco-elasticity, 189
Viscous relative motion, 37
361
AUTHOR INDEX
Bahn, R. C, 227 ,235
Barger, J. E. , 197
Barnes, R. W. , 81
Baxter, B. , 235
Beaver, W. , 267
Birnholz, J. , 287
Bowen , T . , 57
Boyle, D. , 267
Boynard, M. , 165
Brady, J. K. , 19
Busey, H. , 323
Carson, P. L. , 337
Cartensen , E . L . , 29
Christensen, D. , 235
Chivers, R. C. , 343
dayman, W. , 337
Connor, W. G. , 57
Czerwinsk i , M. J. , 303
Dain P., 125
Davidson, C. L. , 179
Dick, D. E. , 337
Dietz, D. R. , 255
Doppman , J . L . , 255
Duck, F. A. , 247
Dunn, F. , 19,43
Edmonds, P. D. , 323
Eggletoti, R. C. , 327
Elbaum, M. E. , 111
Filly, R. A. , 323
Finby, N. , 125
Eraser, J. , 287
Freese, M. , 157
Frizzell , L. A. , 19,43
Fry, E. K. , 85
Fry, F. J. , 85,203
Gaca, A. , 297
Gallagher H. S. , 85
Gammell , P. M. , 101
Glover, G. H., 221
Gore, J. C. , 275
Goss, S. A. , 19,43
Graniiak, R. , 143
Greenleaf , J. F. , 227,235
Halliwell, M., 173
Hanss, M. , 165
Heyser , R. C. , 101
Higgins, F. P. , 255
Hill , C. R. , 247
Holasek, E. , 261
Hunter, L. P. , 143
Jennings, W. D. , 261
Johnson, S. A. , 227,235
Johnston, R. L., 19,43
Joynt, L. , 267
Katz, J, L. , 189
Kessler, L. W. , 73
Keuwez, J. , 121
Kino, G. S. , 287
Kobayashi , T. , 93
Kremkau, F. W. , 81
Kuc, R., 125
Leb, D. E., 303
Le Croissette, D. H. , 101
Lee, P. P. K., 143
Leeman, S. , 275
Lees, S., 179
Lerner, R. M. , 143
Levi, S,, 121
Linzer , M. , 255
Lizzi, F. L., Ill
Lock, E., 297
Lyons, E. A. , 157
Maynard, V. , 19,43
McGraw, C. P. , 81
Metrewel i , C. , 275
Miller, J. G., 37,63
Mimbs, J. W., 63
Mountford, R. A. , 173
Nasoni , R. L. , 57
Nider, L., 43
Norton, S. J. , 255
O'Brien, W. D. , Jr. , 19,43
O'Donnell, M., 37,63
Parkinson, D. B. , 323
Parks, S. I., 255
Parry, R. J. . 343
Phillips, D. J., 209
Pifer, A. E., 57
Plessner, N. J. , 275
Popp, R. , 267
Preston, K. , Jr. , 303
Purnell, E. W., 261
Rakowski, H., 267
Reid, J. M. , 153
Reyes, Z. , 323
Rhyne, T. L., 135
Robinson, D. E. , 281
Rajagopalan, B., 227,235
Roseboro, J. A. , 101
Sanghvi , N. T
Scheiding, W.
Schenk, E. A.
Schwartz, M. ,
Shabason, L. ,
Shawker, T. H
Shideler, R.
Sholes, R. R.
Shung, K. K. ,
Skidmore, R. ,
Skolnik, M. L
Smith, S. W. ,
Sobel, B. E.,
Sollish, B. D
., 85
, 297
, 143
125
337
., 255
W., 255
, 57
153
173
., 303
209
63
., 53
Thomas, P. J., 227
Thurstone, F. L., 209
von Ramm, 0. T. , 209
von Seelen , W. , 297
Waag, R. C, 143
Webb, A. J., 173
Weiss, L. , 317
Wells, P. N. T. , 173
Wessels, G. , 297
Whitcomb, J. A. , 327
Willson, K. , 275
Wilson, R. L. , 101
Woodcock, J. P. , 173
Yoon, H. S., 189
Yuhas, D. E. , 73 ^ •
362
NBS-114A IREV. 0-761
U.S. DEPT. OF COMM.
1. PUBLICATION OR REPORT NO. l||^t||H|||p|P tiOj
BIBLIOGRAPHIC DATA
SHEET
NBS SP 525 p-..*-^r^-->f^-%
li .... .• s
4. TITLE AND SUBTITLE
Ultrasonic Tissue Characterization II
5. Publication Date
April 1979
rmwmi og Orgamnioon Code
7. AUTHOR(S)
Edited by Melvin Linzer
8. Performing Organ, Report No.
9. PERFORMING ORGANIZATION NAME AND ADDRESS
1ft, Pro|ect/Ta«J(/*<Kt( Onii No.
NATIONAL BUREAU OF STANDARDS
DEPARTMENT OF COMMERCE
WASHINGTON, DC 20234
11. Contract/Grant No.
12. SPONSORING ORGANIZATION NAME AND COMPLETE ADDRESS fStreef, city. State. ZIP)
National Science Foundation
Washington, D.C.
and National Institutes of Health
Bethesda, Maryland
13. Type of Report & Period Covered
14. Sponsoring Agency Code
b.^.^.....--........:. :
15. SUPrLEMENTARY NOTES
Library of Congress Catalog Card Number: 79-600026
I I Document describes a computer program; SF-185, FIPS Software Summary, is attached.
16. ABSTRACT (A 200-word or leaa tactual aumrnary of most significant Information. If document includes a significant bibliography or
literature survey, mention it here.)
The Second International *^vmDosium on Ultrasonic Imaging and Tissue
Characterization was held at the National Bureau of Standards on June 13-15, 1977.
The meeting was cosponsored by the National Bureau of Standards, the National
Science Foundation, and the National Institutes of Health. This volume contains
extended and reviewed papers based on 43 of the 53 talks presented at the
Symposium. Topics covered include techniques for measurement of ultrasonic
tissue parameters, the dependence of tissue properties on physical and biological
variables ^e.g., ultrasonic frequency, temperature), mechanisms of ultrasonic
tissue interactions, propagation through bone and skull, tumpr Doppler signatures,
computerized tomography, signal processing and pattern recognition, and tissue
phantoms. A survey of velocity and attenuation data in mammalian tissue is included
in an appendix.
17. KEY WORDS fs/x to twelve entries; alphabetical order; capitalize only the first letter of the first key word unless a proper name;
separated by semicolons)
Absorption; attenuation; computerized tomography; Doppler; impedance; medical
diagnosis; microscopy; pattern recognition; scattering; signal processing;
tissue characterization; tissue parameters; ultrasound; velocity.
18. AVAILABILITY |3 Unlimited
19. SECURITY CLASS
(THIS REPORT)
21. NO. OF
PRINTED PAGES
1 1 For Official Distribution. Do Not Release to NTIS
UNCLASSIFIED
339
rrp Order From Sup. of Doc, Government Printing Office, Washington, DC
^ 20402, SD Stocl< No. SN003-003-0258-3
20. SECURITY CLASS
(THIS PAGE)
22. Price
1 1 Order From National Technical Information Service (NTIS), Springfield,
VA. 22161
UNCLASSIFIED
$5.50
ir U.S. GOVERNMENT PRINTING OFFIfE: 1979 O— 291-026
I
i
I
I
i
i
I
1.
1
i
NBS TECHNICAL PUBLICATIONS
PERIODICALS
JOURNAL OF RESEARCH— The Journal of Research
of the National Bureau of Standards reports NBS research
and development in those disciplines of the physical and
engineering sciences in which the Bureau is active. These
include physics, chemistry, engineering, mathematics, and
computer sciences. Papers cover a broad range of subjects,
with major emphasis on measurement methodology, and
the basic technology underlying standardization. Also in-
cluded from time to time are survey articles on topics closely
related to the Bureau's technical and scientific programs. As
a special service to subscribers each issue contains complete
citations to all recent NBS publications in NBS and non-
NBS media. Issued six times a year. Annual subscription:
domestic $17.00; foreign $21.25. Single copy, $3.00 domestic;
$3.75 foreign.
Note: The Journal was formerly published in two sections:
Section A "Physics and Chemistry" and Section B "Mathe-
matical Sciences."
I DIMENSIONS/NBS
\ This monthly magazine is published to inform scientists,
I engineers, businessmen, industry, teachers, students, and
I consumers of the latest advances in science and technology,
with primary emphasis on the work at NBS. The magazine
[highlights and reviews such issues as energy research, fire
protection, building technology, metric conversion, pollution
,i abatement, health and safety, and consumer product per-
'{formance. In addition, it reports the results of Bureau pro-
grams in measurement standards and techniques, properties
jof matter and materials, engineering standards and services,
jinstrumentation, and automatic data processing.
y Annual subscription: Domestic, $1 1.00; Foreign $13.75
NONPERIODICALS
IMonographs — Major contributions to the technical liter-
ature on various subjects related to the Bureau's scientific
and technical activities.
•Handbooks — Recommended codes of engineering and indus-
trial practice (including safety codes) developed in coopera-
tion with interested industries, professional organizations,
i md regulatory bodies.
jipecial Publications — Include proceedings of conferences
■ponsored by NBS, NBS annual reports, and other special
publications appropriate to this grouping such as wall charts,
TOcket cards, and bibliographies.
Applied Mathematics Series — Mathematical tables, man-
ijials, and studies of special interest to physicists, engineers,
' hemists, biologists, mathematicians, computer programmers,
nd others engaged in scientific and technical work,
''lational Standard Reference Data Series — Provides quanti-
ative data on the physical and chemical properties of
naterials, compiled from the world's literature and critically
valuated. Developed under a world-wide program co-
rdinated by NBS. Program under authority of National
tandard Data Act (Public Law 90-396).
NOTE: At present the principal publication outlet for these
data is the Journal of Physical and Chemical Reference
Data (JPCRD) published quarterly for NBS by the Ameri-
can Chemical Society (ACS) and the American Institute of
Physics (AIP). Subscriptions, reprints, and supplements
available from ACS, 1155 Sixteenth St. N.W., Wash., D.C.
20056.
Building Science Series — Disseminates technical information
developed at the Bureau on building materials, components,
systems, and whole structures. The series presents research
results, test methods, and performance criteria related to the
structural and environmental functions and the durability
and safety characteristics of building elements and systems.
Technical Notes — Studies or reports which are complete in
themselves but restrictive in their treatment of a subject.
Analogous to monographs but not so comprehensive in
scope or definitive in treatment of the subject area. Often
serve as a vehicle for final reports of work performed at
NBS under the sponsorship of other government agencies.
Voluntary Product Standards — Developed under procedures
published by the Department of Commerce in Part 10,
Title 15, of the Code of Federal Regulations. The purpose
of the standards is to establish nationally recognized require-
ments for products, and to provide all concerned interests
with a basis for common understanding of the characteristics
of the products. NBS administers this program as a supple-
ment to the activities of the private sector standardizing
organizations.
Consumer Information Series — Practical information, based
on NBS research and experience, covering areas of interest
to the consumer. Easily understandable language and
illustrations provide useful background knowledge for shop-
ping in today's technological marketplace.
Order above NBS publications from: Superintendent of
Documents, Government Printing Office, Washington, D.C.
20402.
Order following NBS publications — NBSIR's and FIPS from
the National Technical Information Services, Springfield,
Va. 22161.
Federal Information Processing Standards Publications
(FIPS PUB) — Publications in this series collectively consti-
tute the Federal Information Processing Standards Register.
Register serves as the official source of information in the
Federal Government regarding standards issued by NBS
pursuant to the Federal Property and Administrative Serv-
ices Act of 1949 as amended. Public Law 89-306 (79 Stat.
1127), and as implemented by Executive Order 11717
(38 FR 12315, dated May 11, 1973) and Part 6 of Title 15
CFR (Code of Federal Regulations).
NBS Interagency Reports (NBSIR) — ^A special series of
interim or final reports on work performed by NBS for
outside sponsors (both government and non-government).
In general, initial distribution is handled by the sponsor;
public distribution is by the National Technical Information
Services (Springfield, Va. 22161) in paper copy or microfiche
form.
BIBLIOGRAPHIC SUBSCRIPTION SERVICES
e following current-awareness and literature-survey bibli-
aphies are issued periodically by the Bureau:
Cryogenic Data Center Current Awareness Service. A litera-
ture survey issued biweekly. Annual subscription: Domes-
tic, $25.00; Foreign, $30.00.
iquified Natural Gas. A literature survey issued quarterly,
jnnual subscription: $20.00.
Superconducting Devices and Materials. A literature survey
issued quarterly. Annual subscription: $30.00. Send subscrip-
tion orders and remittances for the preceding bibliographic
services to National Bureau of Standards, Cryogenic Data
Center (275.02) Boulder, Colorado 80302.
U.S. DEPARTMENT OF COMMERCE
National Bureau of Standards
Washington. D.C. 20234
OFFICIAL BUSINESS
Penalty for Private Use, S300
POSTAGE AND FEES PAID
U.S. DEPARTMENT OF COMMERCE
COM-219
SPECIAL FOURTH-CLASS RATE
BOOK
i